diff --git a/research/INDEX.md b/research/INDEX.md index 238832b..7cf08c7 100644 --- a/research/INDEX.md +++ b/research/INDEX.md @@ -7,10 +7,10 @@ | Metric | Value | |---|---| | Subjects | 15 | -| Total source documents | 862 | -| — Tier 1 (Source Dossier, independently analyzed) | 266 | -| — Tier 2 (Source Citation Card, listed only) | 596 | -| Total conclusion documents | 34 | +| Total source documents | 921 | +| — Tier 1 (Source Dossier, independently analyzed) | 277 | +| — Tier 2 (Source Citation Card, listed only) | 644 | +| Total conclusion documents | 36 | | Archive opened | 2026-10-09 | | Last updated | 2026-10-25 | @@ -30,7 +30,7 @@ | [`prompt-context-scaling`](prompt-context-scaling/) | 34 (10 / 24) | 3 | 2026-10-18 | 2026-10-18 | Scaling prompts past the context window and reusing RAG/graph for more; tai already matches the converged arsenal, giant prompts should spill to retrievable records (specified, not built), full GraphRAG and memory vendors rejected | | [`background-work-limits`](background-work-limits/) | 10 (4 / 6) | 2 | 2026-10-18 | 2026-10-18 | Bounding parallel subagents and hardening the scheduler; parallel cap 8 with reject-loud, per-tick fires 4, one live run per schedule via lease, stated misfire policy; implemented same day | | [`sandbox-resource-caps`](sandbox-resource-caps/) | 21 (3 / 18) | 1 | 2026-10-18 | 2026-10-18 | Capping the agent sandbox box CPU/memory/pids after the coverage audit found it uncapped; static trio 2g/2.0/256 at create, best-effort update for existing boxes, pids via recreation; implemented same day | -| [`cap-protocol`](cap-protocol/) | 24 (23 / 1) | 2 | 2026-10-18 | 2026-10-18 | Supporting CAP (CLI Agent Protocol) as agent and orchestrator: full read of core + coding profile, quickstart, STATUS, all 5 manifests; selective A2A/ACP wire reads; decided phased rollout without gRPC plus dual A2A spelling | +| [`cap-protocol`](cap-protocol/) | 83 (34 / 49) | 4 | 2026-10-18 | 2026-10-25 | Supporting CAP (CLI Agent Protocol) as agent and orchestrator: full read of core + coding profile, quickstart, STATUS, all 5 manifests; selective A2A/ACP wire reads; decided phased rollout without gRPC plus dual A2A spelling; extended 2026-10-25 with model handling, model-as-tool and skill-management research (ACP config options, MCP sampling, agents-as-tools, advisor tools, subagent and skill docs, RouteLLM), 2 recommendation-only conclusions, nothing built | | [`mcp-client-support`](mcp-client-support/) | 106 (38 / 68) | 3 | 2026-10-18 | 2026-10-18 | Designing tai's MCP client: wire spec (stdio + Streamable HTTP on the 2025-11-25 baseline), pasted-config/GitHub-URL resolver, and the trust/approval model; decided minimal stdlib client, normalizer plus resolver agent, untrusted-until-consent with description pinning, no OAuth in v1, nothing built yet | | [`fuzzy-matching-selectors`](fuzzy-matching-selectors/) | 65 (12 / 53) | 2 | 2026-10-25 | 2026-10-25 | Making `@agent` / `@model` mentions pick the most related candidate: fzf/fzy/Sublime/clangd subsequence scoring, Damerau-Levenshtein, pg_trgm, RapidFuzz WRatio, difflib, CPython suggestions surveyed; decided one generic `FuzzyMatcher` (four signals, deterministic ties, floor 0.3 across agents and models) and implemented it the same day | @@ -63,6 +63,7 @@ | 2026-10-18 | `cap-protocol` | New subject: 24 sources (23 Tier 1 / 1 Tier 2) + 2 conclusions, covering CAP core v1 and the coding profile read in full, quickstart, STATUS, all 5 example manifests, A2A spec + proto (selective), 10 ACP pages (selective), and the reference A2A driver (grepped). Decisions: implement all bindings except gRPC in 5 phases (manifest/events, agent speak, PTY driver, stream-json + orchestration, ACP/A2A + auto-answer); try draft then current A2A endpoint spellings. Counts verified directly against the written files via `research/audit.py` (PASS). | | 2026-10-18 | `mcp-client-support` | New subject: 106 sources (38 Tier 1 / 68 Tier 2) + 3 conclusions, covering the MCP wire spec (lifecycle, transports, tools, utilities, 2024-11-05 legacy, versioning, changelog), client config ecosystems (Claude Code, VS Code, Cursor, Codex TOML, mcp-remote, Smithery, registries, Glama), and security/operations (authorization draft, trust model, poisoning, rug pulls, production ops) via 3 gather subagents plus first-hand fetches of the load-bearing spec, client-config, and OWASP pages. Decisions: minimal stdlib client on the 2025-11-25 baseline (stdio + Streamable HTTP, no SDK, no OAuth in v1); deterministic normalizer plus resolver subagent for pasted configs and GitHub URLs with live-handshake verification; untrusted-until-consent trust model with description-hash pinning and approval tiers. Nothing built yet. Counts verified directly against the written files via `research/audit.py` (PASS). | | 2026-10-25 | `fuzzy-matching-selectors` | New subject: 65 sources (12 Tier 1 / 53 Tier 2) + 2 conclusions, from 10 searches and 14 fetches on fuzzy string matching (fzf algo.go and man page, fzy ALGORITHM.md and match.c, Sublime fuzzy match via an AsciiDoc copy because the original blog returned 403, clangd FuzzyMatch, pg_trgm docs, RapidFuzz WRatio source, Jaro-Winkler and Damerau-Levenshtein articles, difflib docs, CPython traceback suggestions). Decisions: one generic `detz::FuzzyMatcher` (subsequence DP with fzf constants + restricted Damerau-Levenshtein + per-token typo + trigram Jaccard, deterministic shorter-then-earlier ties) and a fuzzy catalog resolution order with floor 0.3. Implemented the same session in taicpp (src/util/fuzzy.*, src/cap/cap_model_catalog.cpp, tests/fuzzy_test.cpp). Carried gaps: VS Code fuzzyScorer.ts never retrieved; fzy constants not retrieved (fzf's used); no measurement on a real model-id corpus; the Jaro-Winkler prefix-inflation claim rests on a search listing only. | +| 2026-10-25 | `cap-protocol` | Extended with 59 sources (11 Tier 1 / 48 Tier 2) + 2 conclusions (`cap-model-handling-practice`, `agent-delegation-tool-plus-skill`), from 10 searches and 15 fetches on how CAP and neighboring protocols treat models and cost, and whether delegation should be a tool plus a skill. Both conclusions are recommendations pending retoor, not decisions (their Decided by field says so). Two existing dossiers (`cap-core-v1`, `cap-profile-coding-v1`) were forward-corrected with model-handling findings. Carried gaps: Claude Code sub-agent and skills pages read only to character 100000; advisor benefit rests on one unverified transcript; the A2A searches found no model metadata field; CAP STATUS and the OpenAI handoffs page were not re-read. Nothing built. | ## How to use this index diff --git a/research/cap-protocol/agentclientprotocol-rfd-model-config-category.md b/research/cap-protocol/agentclientprotocol-rfd-model-config-category.md new file mode 100644 index 0000000..59be58c --- /dev/null +++ b/research/cap-protocol/agentclientprotocol-rfd-model-config-category.md @@ -0,0 +1,28 @@ +# ACP RFD: model config category + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | ACP RFD: model config category | +| Source URL | https://agentclientprotocol.com/rfds/model-config-category.md | +| Source type | standard | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the ACP model-selection search; proposes a `model_config` category grouping settings next to the model picker. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/agentclientprotocol-session-config-options.md b/research/cap-protocol/agentclientprotocol-session-config-options.md new file mode 100644 index 0000000..098c338 --- /dev/null +++ b/research/cap-protocol/agentclientprotocol-session-config-options.md @@ -0,0 +1,52 @@ +# ACP: Session Config Options + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | ACP: Session Config Options | +| Source URL | https://agentclientprotocol.com/protocol/session-config-options | +| Source publisher / author | Agent Client Protocol project | +| Source type | standard | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +The ACP mechanism by which an agent advertises per-session selectors (model, mode, reasoning level) and a client changes them. It is the closest existing standard to 'list an agent's models and pick one'. + +## Key findings + +- Each config option has `id`, `name`, optional `description`, optional `category`, `type` (`select` by default, `boolean` only if the client advertised support), `currentValue`, and `options` (flat values or groups, not mixed). +- Reserved categories: `mode`, `model`, `model_config`, `thought_level`; names starting with `_` are custom; clients MUST handle missing or unknown categories gracefully. +- The agent MAY return `configOptions` during session setup; the client sets one with `session/set_config_option` (`sessionId`, `configId`, `value`) and the agent MUST answer with the complete list of options and values. +- The agent announces changes with a `session/update` of `sessionUpdate: config_option_update` carrying the full `configOptions` array. +- Agents MUST always provide a default for every option; a client that meets an unknown `type` SHOULD ignore that option and the agent keeps its default. +- Order is significant: higher-priority options first; clients use the order to break ties within a category. +- If an agent provides `configOptions`, clients SHOULD use them instead of `modes`, which will be removed. + +## Pros / strengths + +Normative (MUST/SHOULD) and small; gives a live, per-agent, per-session model list instead of a static one, with a defined default and ordering. + +## Cons / limitations / caveats + +Only exists for agents that speak ACP; the page does not say how large option lists should be capped; the `model_config` category is separate RFD territory. + +## Relevance to tai + +tai's catalog reads a manifest list or a `--cap-models` probe, both static. For `acp_stdio` agents the live truth is the `model`-category config option from session setup. Recommended as the first discovery source for that binding, with ACP's default-on-unknown rule as the error-resistance model. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/agentclientprotocol-v2-session-config-options.md b/research/cap-protocol/agentclientprotocol-v2-session-config-options.md new file mode 100644 index 0000000..d95c382 --- /dev/null +++ b/research/cap-protocol/agentclientprotocol-v2-session-config-options.md @@ -0,0 +1,28 @@ +# ACP v2: Session Config Options + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | ACP v2: Session Config Options | +| Source URL | https://agentclientprotocol.com/protocol/v2/session-config-options | +| Source type | standard | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the ACP model-selection search; the v2 copy of the page read as Tier 1. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/agentpatterns-agent-as-tool-vs-handoff.md b/research/cap-protocol/agentpatterns-agent-as-tool-vs-handoff.md new file mode 100644 index 0000000..94babff --- /dev/null +++ b/research/cap-protocol/agentpatterns-agent-as-tool-vs-handoff.md @@ -0,0 +1,28 @@ +# Agent as Tool vs Handoff: Who Keeps the Conversation + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent as Tool vs Handoff: Who Keeps the Conversation | +| Source URL | https://agentpatterns.ai/patterns/agent-design/agent-as-tool-vs-handoff/ | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search; claims Claude Code ships only the tool form, not verified. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/agentskills-io-specification.md b/research/cap-protocol/agentskills-io-specification.md new file mode 100644 index 0000000..af6a5e3 --- /dev/null +++ b/research/cap-protocol/agentskills-io-specification.md @@ -0,0 +1,49 @@ +# Agent Skills specification + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | Agent Skills specification | +| Source URL | https://agentskills.io/specification | +| Source publisher / author | Agent Skills project | +| Source type | standard | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +The open SKILL.md format that tai's skill system already follows. + +## Key findings + +- A skill is a directory with `SKILL.md` (YAML frontmatter plus Markdown) and optional `scripts/`, `references/`, `assets/`. +- `name`: 1-64 chars, lowercase alphanumerics and hyphens, no leading, trailing or consecutive hyphens, must match the directory name. `description`: 1-1024 chars, should say what the skill does and when to use it. Optional `license`, `compatibility` (max 500), `metadata`, experimental `allowed-tools`. +- Progressive disclosure: metadata about 100 tokens loaded at startup for all skills; instructions under 5000 tokens recommended on activation; resources on demand. +- Keep `SKILL.md` under 500 lines; keep file references one level deep from `SKILL.md`; validate with `skills-ref validate`. + +## Pros / strengths + +Normative limits that a bundled tai skill can be checked against. + +## Cons / limitations / caveats + +Describes format, not when a skill beats a tool. + +## Relevance to tai + +Constraints for a bundled `cap-agents` skill: single-line description under 1024 chars (tai's own frontmatter subset needs single-line values), body under 5000 tokens, detail in `references/`. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/aguidetocloud-openai-agents-concepts.md b/research/cap-protocol/aguidetocloud-openai-agents-concepts.md new file mode 100644 index 0000000..e954964 --- /dev/null +++ b/research/cap-protocol/aguidetocloud-openai-agents-concepts.md @@ -0,0 +1,28 @@ +# Core concepts: agents, handoffs, tools, guardrails + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Core concepts: agents, handoffs, tools, guardrails | +| Source URL | https://claw.aguidetocloud.com/openai/agents-sdk/concepts/ | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/aitmpl-sonnet-opus-advisor-haiku-subagents.md b/research/cap-protocol/aitmpl-sonnet-opus-advisor-haiku-subagents.md new file mode 100644 index 0000000..254c52c --- /dev/null +++ b/research/cap-protocol/aitmpl-sonnet-opus-advisor-haiku-subagents.md @@ -0,0 +1,28 @@ +# Sonnet, Opus advisor, Haiku subagents setting + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Sonnet, Opus advisor, Haiku subagents setting | +| Source URL | https://www.aitmpl.com/component/settings/model/sonnet-opus-advisor-haiku-subagents | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search; a settings template combining the three. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/algolia-agentskills-repo-docs.md b/research/cap-protocol/algolia-agentskills-repo-docs.md new file mode 100644 index 0000000..6cb9ce4 --- /dev/null +++ b/research/cap-protocol/algolia-agentskills-repo-docs.md @@ -0,0 +1,28 @@ +# Agent Skills repository docs (Algolia index) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent Skills repository docs (Algolia index) | +| Source URL | https://docsearch.algolia.com/mcp/docs/repo/agentskills/agentskills | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/aman-ai-agent-skills-primer.md b/research/cap-protocol/aman-ai-agent-skills-primer.md new file mode 100644 index 0000000..b5f80f4 --- /dev/null +++ b/research/cap-protocol/aman-ai-agent-skills-primer.md @@ -0,0 +1,28 @@ +# Primer: Agent Skills (aman.ai) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Primer: Agent Skills (aman.ai) | +| Source URL | https://aman.ai/primers/ai/agent-skills/ | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/anthropic-multi-agent-research-system.md b/research/cap-protocol/anthropic-multi-agent-research-system.md new file mode 100644 index 0000000..0b6a81f --- /dev/null +++ b/research/cap-protocol/anthropic-multi-agent-research-system.md @@ -0,0 +1,53 @@ +# How we built our multi-agent research system (Anthropic Engineering) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | How we built our multi-agent research system (Anthropic Engineering) | +| Source URL | https://www.anthropic.com/engineering/multi-agent-research-system | +| Source publisher / author | Anthropic | +| Source type | blog post | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +Anthropic's account of an orchestrator-worker research system: a lead agent plans and spawns parallel subagents with their own context windows. Includes numbers on performance and token cost, prompting principles for delegation, and when the pattern does not fit. + +## Key findings + +- Lead Claude Opus 4 with Sonnet 4 subagents beat single-agent Opus 4 by 90.2% on the internal research eval. +- On BrowseComp three factors explained 95% of performance variance; token usage alone explained 80%, with tool calls and model choice the others. +- Agents use about 4x the tokens of chat; multi-agent systems about 15x. Parallelization cut research time by up to 90% on complex queries. +- Delegation must be taught: each subagent needs an objective, an output format, tool and source guidance, and clear task boundaries; vague instructions caused duplicated work. +- Scale effort to complexity: 1 agent with 3-10 tool calls for simple facts, 2-4 subagents with 10-15 calls for comparisons, 10+ subagents for complex research. +- Does not fit: tasks where all agents share the same context or have many dependencies, most coding tasks (fewer parallelizable parts), weak real-time coordination, and low-value tasks that cannot pay the token cost. +- Production: agents are stateful so errors compound (resume from checkpoints, tell agents when a tool fails); leads wait synchronously for each subagent batch; start evals with about 20 real queries; judge end state for stateful agents. +- Upgrading to Sonnet 4 gave a larger gain than doubling the token budget on Sonnet 3.7. + +## Pros / strengths + +Primary source with concrete numbers, and an honest list of where delegation fails, including coding. + +## Cons / limitations / caveats + +A research workload, not an interactive coding agent; numbers come from Anthropic's internal eval and BrowseComp, not independently reproduced. 15x is a ballpark for their system. + +## Relevance to tai + +Direct evidence for two rules: delegation to another model costs a multiple of the tokens (so cap it and count it), and a delegation brief needs objective, format, tools and boundaries (so a delegation tool needs a structured input, not a free string). + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/arxiv-2406-18665-routellm.md b/research/cap-protocol/arxiv-2406-18665-routellm.md new file mode 100644 index 0000000..b39811d --- /dev/null +++ b/research/cap-protocol/arxiv-2406-18665-routellm.md @@ -0,0 +1,28 @@ +# RouteLLM: Learning to Route LLMs with Preference Data + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | RouteLLM: Learning to Route LLMs with Preference Data | +| Source URL | https://arxiv.org/html/2406.18665v2 | +| Source type | paper | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search; the paper behind the LMSYS post, not opened. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/berkeley-sky-routellm.md b/research/cap-protocol/berkeley-sky-routellm.md new file mode 100644 index 0000000..27bb9cf --- /dev/null +++ b/research/cap-protocol/berkeley-sky-routellm.md @@ -0,0 +1,28 @@ +# RouteLLM project page (Berkeley Sky) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | RouteLLM project page (Berkeley Sky) | +| Source URL | https://sky.cs.berkeley.edu/project/routellm/ | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/bytebytego-anthropic-multi-agent.md b/research/cap-protocol/bytebytego-anthropic-multi-agent.md new file mode 100644 index 0000000..4b36a9f --- /dev/null +++ b/research/cap-protocol/bytebytego-anthropic-multi-agent.md @@ -0,0 +1,28 @@ +# How Anthropic Built a Multi-Agent Research System (ByteByteGo) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | How Anthropic Built a Multi-Agent Research System (ByteByteGo) | +| Source URL | https://blog.bytebytego.com/p/how-anthropic-built-a-multi-agent | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the multi-agent search; secondary summary, source of the 15x figure in listings. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/cap-core-v1.md b/research/cap-protocol/cap-core-v1.md index 37c02bc..8580382 100644 --- a/research/cap-protocol/cap-core-v1.md +++ b/research/cap-protocol/cap-core-v1.md @@ -32,6 +32,9 @@ The normative core: PTY is the REQUIRED universal binding, four OPTIONAL fast-pa - Reverse RPC has no defined wire encoding for stream-json/gRPC bindings; only the ACP binding gives it a native transport (JSON-RPC methods). - Error registry: -32001..-32010 with string twins; `cap.error` string codes include `rate_limit`, `pty_died`, `parse_failed`, `context_overflow`. +- (added 2026-10-25, re-read for model handling) `cap.session.config` carries a `model` field (example `claude-opus-4-7`, allowed values undefined) and `budget_usd`; the driver applies `model` before spawning or after a binding-specific handshake (section 3.3). The manifest's only model field is `startup.model_arg`; there is no `models` list and no `effort` field. +- (added 2026-10-25) No method or event switches models mid-session and the spec defines no model-selection or routing strategy. `cap.usage` reports `model_id`, token counts and `cost_usd_estimate`; the orchestrator MUST aggregate usage and enforce budgets by `cap.cancel` ('Sub-agents MUST NOT be trusted to self-enforce budget', section 10.5). Errors: `model_unavailable`, `-32004 cap_budget_exceeded`. + ## Pros / strengths - Complete contract in one file: events, lifecycle, orchestration, errors, and informative binding mappings (C.1-C.4) sufficient to implement drivers without guessing. @@ -54,3 +57,4 @@ The build blueprint for both directions: tai-as-agent implements §3.1 via `--ca | Date | Change | |---|---| | 2026-10-18 | Initial dossier created. | +| 2026-10-25 | Forward correction: model-handling findings added from a re-read; the earlier note on routing strategies (static, llm, hybrid) comes from the quickstart, not from this document. | diff --git a/research/cap-protocol/cap-profile-coding-v1.md b/research/cap-protocol/cap-profile-coding-v1.md index 96a9341..6f71bdc 100644 --- a/research/cap-protocol/cap-profile-coding-v1.md +++ b/research/cap-protocol/cap-profile-coding-v1.md @@ -32,6 +32,8 @@ The first vertical: `[capabilities.coding]` manifest section, 6 fs reverse-RPC m - ACP bridge table (§9): fs.list/exists/delete have NO ACP counterpart and go over `_cap/` extension methods. - Multi-agent worktree naming suggestion: `/.cap-worktrees//` (the reference CLI instead uses `.cap//`). +- (added 2026-10-25, re-read for model handling) The profile defines no model selection, cost, budget or effort; the only model mention is an example `model` field in a transcript artifact. Per-agent settings an orchestrator does set: `tool_permission` (`none`, `confirm`, `interactive`), `tools_allowed` / `tools_denied`, `fs.scope`, `terminal.concurrent_max`; sub-agents SHOULD each get their own git worktree and staged diffs SHOULD be vetted before a commit. + ## Pros / strengths - Tight, implementable RPC surface: every method has params/result/errors, and the scope table leaves no enforcement ambiguity. @@ -54,3 +56,4 @@ Defines tai's coding-profile surface both ways: orchestrator MUST implement the | Date | Change | |---|---| | 2026-10-18 | Initial dossier created. | +| 2026-10-25 | Forward correction: model-handling findings added from a re-read. | diff --git a/research/cap-protocol/claude-code-advisor-docs.md b/research/cap-protocol/claude-code-advisor-docs.md new file mode 100644 index 0000000..7a2aadf --- /dev/null +++ b/research/cap-protocol/claude-code-advisor-docs.md @@ -0,0 +1,52 @@ +# Claude Code: Escalate hard decisions with the advisor tool + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | Claude Code: Escalate hard decisions with the advisor tool | +| Source URL | https://code.claude.com/docs/en/advisor | +| Source publisher / author | Anthropic | +| Source type | vendor docs | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +Documentation of the advisor tool: the main model consults a second, typically stronger model at decision points and applies the guidance. + +## Key findings + +- Claude decides when to call the advisor (before committing to an approach, when an error recurs, before declaring done); the advisor receives the full conversation including tool calls and results. +- It is a server tool, experimental, Anthropic API only; the user picks the advisor model (`/advisor`, `advisorModel`, `--advisor`); an advisor must rank at or above the main model. +- Cost: advisor tokens bill at the advisor model's rates on top of the main model's; there is no setting to cap or force advisor calls. +- The advisor adds less value on short tasks or when every turn needs the strongest model; for those, switch the main model instead. +- Claude does not follow advice blindly: when its own evidence contradicts a claim it surfaces the conflict. +- Comparison table: advisor (decision points, model-called), `opusplan` (plan mode), subagents with `model` (whole subtask), `/model` (from the next request). +- Subagents inherit the configured advisor and the same pairing check applies to their own model. + +## Pros / strengths + +Shows model-as-tool shipped in a major product, with the failure modes (unavailable, declined) surfaced to the user. + +## Cons / limitations / caveats + +Experimental, vendor-specific, and the pairing table is tied to Claude models. No cap on calls. + +## Relevance to tai + +Concrete model-as-tool precedent and the comparison of four ways to combine models. The lack of a call cap is a gap tai should close with a static ceiling. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/claude-code-skills-docs.md b/research/cap-protocol/claude-code-skills-docs.md new file mode 100644 index 0000000..f2f5e28 --- /dev/null +++ b/research/cap-protocol/claude-code-skills-docs.md @@ -0,0 +1,51 @@ +# Claude Code: Skills + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | Claude Code: Skills | +| Source URL | https://code.claude.com/docs/en/skills | +| Source publisher / author | Anthropic | +| Source type | vendor docs | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | Medium | +| Status | Active | + +## Summary + +Claude Code's skills documentation: loading, frontmatter, model pinning, forked subagent context and size guidance. Only the first 100000 of 118114 characters were read. + +## Key findings + +- Skill descriptions are loaded into context so the model knows what exists; the body loads only when invoked, so long reference material 'costs almost nothing until you need it'. +- Frontmatter includes `description`, `when_to_use`, `disable-model-invocation`, `user-invocable`, `allowed-tools`, `model` (applies for the rest of the current turn), `effort`, `context: fork` (run in an isolated subagent), `agent`, `background`. +- Combined `description` plus `when_to_use` is truncated at 1,536 characters in the listing; keep the body under 500 lines; after compaction re-attached skills share a 25,000-token budget and keep their first 5,000 tokens. +- `context: fork` skills do not see the conversation history, so their instructions must stand on their own. +- Create a skill 'when you keep pasting the same instructions, checklist, or multi-step procedure into chat'. +- Custom commands have been merged into skills. + +## Pros / strengths + +Shows a skill can both carry guidance and pin the model or fork a subagent for its own run. + +## Cons / limitations / caveats + +Tail unread; some fields are product-specific and not part of the open Agent Skills format. + +## Relevance to tai + +Supports the split 'tool does the action, skill teaches when and how': a skill carries a recurring procedure at near-zero idle cost, but enforces nothing by itself. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/claude-code-sub-agents-docs.md b/research/cap-protocol/claude-code-sub-agents-docs.md new file mode 100644 index 0000000..8261e7e --- /dev/null +++ b/research/cap-protocol/claude-code-sub-agents-docs.md @@ -0,0 +1,51 @@ +# Claude Code: Custom subagents + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | Claude Code: Custom subagents | +| Source URL | https://code.claude.com/docs/en/sub-agents | +| Source publisher / author | Anthropic | +| Source type | vendor docs | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | Medium | +| Status | Active | + +## Summary + +Claude Code's documentation of subagents: file format, the `model` field, model resolution order, delegation modes and context isolation. Only the first 100000 of 116349 characters were read. + +## Key findings + +- Frontmatter: `name`, `description` (when to delegate), `tools` (inherits all if omitted), `model` (`sonnet`, `opus`, `haiku`, `fable`, a full id, or `inherit`). +- Model resolution order: per-invocation `model` parameter, then the definition's `model`, then the `CLAUDE_CODE_SUBAGENT_MODEL` environment variable, then the main conversation's model; before v2.1.251 the environment variable came first. +- Delegation is automatic from descriptions (include 'use proactively' to encourage it), by natural language, by @-mention (guarantees the run), or session-wide. +- Each subagent has its own context window, system prompt, tool access and permissions; a non-fork subagent does not see the conversation history and works from a delegation message; only a summary returns. +- Subagent requests count toward the same usage limits as the main conversation; costs are controlled by routing tasks to faster, lower-cost models like Haiku. +- Combined subagent descriptions above 15,000 tokens trigger a startup warning; use the main conversation for frequent back-and-forth or quick targeted changes. + +## Pros / strengths + +A shipped product's explicit, ordered precedence for the model of a delegated worker, and a cost-routing rationale. + +## Cons / limitations / caveats + +Unread tail of the page; the precedence change between versions shows the rule is unstable across releases. + +## Relevance to tai + +Precedent for a tai rule: per-call model beats the agent's configured model beats a global default beats the parent's model. Also shows 'description decides delegation' costs context, which argues for few, well-described delegation tools. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/claude-zero-trust-current-threats.md b/research/cap-protocol/claude-zero-trust-current-threats.md new file mode 100644 index 0000000..57e2300 --- /dev/null +++ b/research/cap-protocol/claude-zero-trust-current-threats.md @@ -0,0 +1,28 @@ +# Zero trust for AI agents: current threats to agentic systems + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Zero trust for AI agents: current threats to agentic systems | +| Source URL | https://claude.com/resources/guides/zero-trust-for-ai-agents/current-threats-to-agentic-systems | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the delegation-security search; names unscoped manager-to-worker delegation and low-to-high privilege relay. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/conclusions/agent-delegation-tool-plus-skill.md b/research/cap-protocol/conclusions/agent-delegation-tool-plus-skill.md new file mode 100644 index 0000000..eb2917c --- /dev/null +++ b/research/cap-protocol/conclusions/agent-delegation-tool-plus-skill.md @@ -0,0 +1,57 @@ +# Conclusion: Model and agent delegation as a bounded tool, guided by a skill + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Conclusion | +| Tier | N/A | +| Date decided | 2026-10-25 (recommendation only, retoor has not yet decided) | +| Decided by | tai (agent, Claude Sonnet 5) as a recommendation, pending retoor | +| Status | Active | +| Confidence | Medium | + +## Question / problem statement + +Should tai let its own model call other CAP agents or models as a tool (model-as-tool), and should that capability be managed through a skill. + +## Conclusion + +Yes to a bounded agent-as-tool, and yes to a skill, with different jobs. A tool `ask_agent` (structured input: agent, optional model hint, prompt; returns only the delegate's final text) does the action: nested run in its own context, delegate permissions scoped and confirm-by-default, results treated as untrusted data, a depth of 1 (the tool is removed inside a delegated run), a static per-turn ceiling, and cost attributed to the delegate's model. It reuses `CapMention`. A single-turn advisor-style `ask_model` (no tools, one answer) is a possible second tool, deferred until delegation has been used. A bundled skill (for example `cap-agents`) carries the judgment: when delegation pays, how to write a brief (objective, output format, tool and source guidance, boundaries), how to pick a model tier, how to read the failure report. A skill cannot enforce caps or approvals, so it must not be the only mechanism. The human-chosen `@agent` route stays as the handoff form. + +## Supporting sources + +- [openai-agents-python-tools-agent-as-tool](../openai-agents-python-tools-agent-as-tool.md): shape of an agent-as-tool (own context, final output only, enable switch, approval pause). +- [openai-agents-python-multi-agent-orchestration](../openai-agents-python-multi-agent-orchestration.md): tool form when the manager keeps the reply. +- [anthropic-multi-agent-research-system](../anthropic-multi-agent-research-system.md): token multiple, brief contents, poor fit for most coding. +- [claude-code-advisor-docs](../claude-code-advisor-docs.md) and [openrouter-advisor-server-tool](../openrouter-advisor-server-tool.md): advisor form, depth guard, per-request cap, separate billing. +- [claude-code-skills-docs](../claude-code-skills-docs.md) and [agentskills-io-specification](../agentskills-io-specification.md): skills as low idle-cost guidance with size limits. +- [claude-code-sub-agents-docs](../claude-code-sub-agents-docs.md): description-driven delegation and its context cost. +- Tier 2: the Anthropic zero-trust guide and confused-deputy commentary back the least-privilege and untrusted-output rules. + +## Alternatives considered and rejected + +- Skill only: cheap and quick, but it enforces nothing and depends on the model following prose. +- Tool only: enforces limits but gives the model no guidance on when delegation pays, which the evidence says is where delegation fails (vague briefs, duplicated work). +- Automatic delegation from descriptions (Claude Code style): a bigger surface and a context cost for every described agent; tai has about 60 tools already and keyword-selects them. +- Handoff-only (the current `@agent` route): fine for humans, but the model cannot use another agent mid-task. +- Advisor first: its cited gain rests on one unverified transcript and tai's free backends are mostly one tier, so the benefit is unmeasured. + +## Risks / open questions + +- Anthropic reports most coding work is poorly parallelizable and delegation costs about 15x tokens; the tool could be net-negative on small tasks, so the skill must say when not to delegate. +- Delegated agents run with their own tools; the confused-deputy risk is real and the scoping is only as good as each agent's permission mode. +- A bundled skill must be added to tai.py first because the C++ port regenerates its bundled skills from it. +- No measurement exists on tai workloads; a trial with a counter on delegations and their cost should precede widening. + +## Action taken + +Recommendation only. Nothing built; retoor has not decided. If approved, the build order is `ask_agent` tool with caps, then the `cap-agents` skill in tai.py and the regenerated bundle, then evaluate before considering an advisor tool. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial conclusion recorded as a recommendation. | diff --git a/research/cap-protocol/conclusions/cap-model-handling-practice.md b/research/cap-protocol/conclusions/cap-model-handling-practice.md new file mode 100644 index 0000000..e10d18b --- /dev/null +++ b/research/cap-protocol/conclusions/cap-model-handling-practice.md @@ -0,0 +1,54 @@ +# Conclusion: How tai should handle models of CAP agents + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Conclusion | +| Tier | N/A | +| Date decided | 2026-10-25 (recommendation only, retoor has not yet decided) | +| Decided by | tai (agent, Claude Sonnet 5) as a recommendation, pending retoor | +| Status | Active | +| Confidence | Medium | + +## Question / problem statement + +What do CAP and the neighboring protocols say about models and their usage, and what should tai's CAP side do about discovery, selection, precedence, cost and budgets. + +## Conclusion + +Treat a model as session configuration attached to an agent, resolved by an explicit precedence, never as agent-private knowledge. (1) Discovery order: the live `model`-category config option from ACP session setup for `acp_stdio` agents, then the manifest `[models]` list, then the `--cap-models` probe (CAP itself has no model list). (2) Selection: the caller passes a name or substring hint and the side that owns the catalog resolves it (MCP sampling hints, Claude Code aliases), which tai's fuzzy matcher already does. (3) Precedence for a delegated call: per-call model, then the agent's configured model, then a global default, then the parent's model; document it, because Claude Code's own order changed between releases. (4) Pass the model through `model_arg`; when an agent has none, warn instead of failing (already done). (5) Count cost from `cap.usage` (`model_id`, tokens, `cost_usd_estimate`) and enforce budgets in the orchestrator, never in the agent. (6) Expect delegation to cost a multiple of chat tokens (Anthropic reports about 15x) and cap it. + +## Supporting sources + +- [cap-core-v1](../cap-core-v1.md): `model` in session config, `model_arg`, `cap.usage`, orchestrator-enforced budgets (forward-corrected in this pass). +- [cap-profile-coding-v1](../cap-profile-coding-v1.md): no model or cost fields; per-agent permission and scope settings. +- [agentclientprotocol-session-config-options](../agentclientprotocol-session-config-options.md): live model options, defaults, ordering. +- [modelcontextprotocol-sampling-model-preferences](../modelcontextprotocol-sampling-model-preferences.md): hints plus priorities, owner decides. +- [claude-code-sub-agents-docs](../claude-code-sub-agents-docs.md): per-worker model and precedence. +- [anthropic-multi-agent-research-system](../anthropic-multi-agent-research-system.md): token multiple and model-tier results. +- [lmsys-routellm-blog](../lmsys-routellm-blog.md): two-tier routing savings. + +## Alternatives considered and rejected + +- Wait for CAP to standardize a model list: CAP's own STATUS marks no such work, and tai needs it now. +- A trained router (RouteLLM style): needs training data and a dependency; the calling model or the user already knows when a task is hard. +- Per-agent hard-coded model tables in tai: stale within weeks as providers rename models. + +## Risks / open questions + +- ACP discovery is not implemented in tai's CAP code; only the manifest list and probe exist. +- The precedence order is a recommendation, not something users have been asked about. +- Cost fields depend on agents reporting `cost.metered`; unmetered agents give no budget signal. + +## Action taken + +Nothing built beyond what already exists (manifest list, probe, fuzzy resolution). ACP discovery and the precedence rule are specified here only. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial conclusion recorded as a recommendation. | diff --git a/research/cap-protocol/corti-agent-cards.md b/research/cap-protocol/corti-agent-cards.md new file mode 100644 index 0000000..6dcfff1 --- /dev/null +++ b/research/cap-protocol/corti-agent-cards.md @@ -0,0 +1,28 @@ +# Agent cards (Corti docs) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent cards (Corti docs) | +| Source URL | https://docs.corti.ai/agentic/agent-cards.md | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the A2A search; no model metadata field found in any listing. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/csharpcorner-openai-agents-sdk.md b/research/cap-protocol/csharpcorner-openai-agents-sdk.md new file mode 100644 index 0000000..30557ce --- /dev/null +++ b/research/cap-protocol/csharpcorner-openai-agents-sdk.md @@ -0,0 +1,28 @@ +# OpenAI Agents SDK overview (C# Corner) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | OpenAI Agents SDK overview (C# Corner) | +| Source URL | https://www.c-sharpcorner.com/article/openai-agents-sdk-what-it-is-and-how-to-build-production-agents/ | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/datacamp-agent-skills.md b/research/cap-protocol/datacamp-agent-skills.md new file mode 100644 index 0000000..14af122 --- /dev/null +++ b/research/cap-protocol/datacamp-agent-skills.md @@ -0,0 +1,28 @@ +# Agent Skills (DataCamp) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent Skills (DataCamp) | +| Source URL | https://www.datacamp.com/nl/blog/agent-skills | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/dzone-progressive-disclosure-ai-agents.md b/research/cap-protocol/dzone-progressive-disclosure-ai-agents.md new file mode 100644 index 0000000..ffd6ca2 --- /dev/null +++ b/research/cap-protocol/dzone-progressive-disclosure-ai-agents.md @@ -0,0 +1,28 @@ +# Progressive disclosure for AI agents (DZone) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Progressive disclosure for AI agents (DZone) | +| Source URL | https://dzone.com/articles/progressive-disclosure-ai-agents | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/gitconnected-orchestrator-worker-pattern.md b/research/cap-protocol/gitconnected-orchestrator-worker-pattern.md new file mode 100644 index 0000000..46d1df7 --- /dev/null +++ b/research/cap-protocol/gitconnected-orchestrator-worker-pattern.md @@ -0,0 +1,28 @@ +# The orchestrator-worker pattern + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | The orchestrator-worker pattern | +| Source URL | https://levelup.gitconnected.com/the-orchestrator-worker-pattern-how-to-structure-a-multi-agent-system-58d469a8e5e1 | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the multi-agent search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/github-openai-agents-js-pr-455.md b/research/cap-protocol/github-openai-agents-js-pr-455.md new file mode 100644 index 0000000..e83754b --- /dev/null +++ b/research/cap-protocol/github-openai-agents-js-pr-455.md @@ -0,0 +1,28 @@ +# openai-agents-js pull request 455 + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | openai-agents-js pull request 455 | +| Source URL | https://github.com/openai/openai-agents-js/pull/455/files | +| Source type | repository | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search; quotes the manager pattern from the JS docs. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/horadecodar-haiku-subagentes.md b/research/cap-protocol/horadecodar-haiku-subagentes.md new file mode 100644 index 0000000..ec4eaaf --- /dev/null +++ b/research/cap-protocol/horadecodar-haiku-subagentes.md @@ -0,0 +1,28 @@ +# Quando vale a pena usar Haiku nos subagentes do Claude Code? + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Quando vale a pena usar Haiku nos subagentes do Claude Code? | +| Source URL | https://horadecodar.com.br/?p=43584 | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search; warns against pinning Haiku to deciding roles. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/itlibra-claude-code-subagent-model.md b/research/cap-protocol/itlibra-claude-code-subagent-model.md new file mode 100644 index 0000000..0bc1b9c --- /dev/null +++ b/research/cap-protocol/itlibra-claude-code-subagent-model.md @@ -0,0 +1,28 @@ +# Claude Code subagent model (itlibra) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Claude Code subagent model (itlibra) | +| Source URL | https://arte.itlibra.com/en/articles/claude-code-subagent-model | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/joinleland-claude-subagents.md b/research/cap-protocol/joinleland-claude-subagents.md new file mode 100644 index 0000000..844da3c --- /dev/null +++ b/research/cap-protocol/joinleland-claude-subagents.md @@ -0,0 +1,28 @@ +# Claude Code Subagents vs. Agents + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Claude Code Subagents vs. Agents | +| Source URL | https://www.joinleland.com/library/a/claude-subagents | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/jsr-mcpc-acp-ai-provider-languagemodel.md b/research/cap-protocol/jsr-mcpc-acp-ai-provider-languagemodel.md new file mode 100644 index 0000000..ac3043e --- /dev/null +++ b/research/cap-protocol/jsr-mcpc-acp-ai-provider-languagemodel.md @@ -0,0 +1,28 @@ +# ACPLanguageModel (acp-ai-provider) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | ACPLanguageModel (acp-ai-provider) | +| Source URL | https://jsr.io/@mcpc/acp-ai-provider/doc/~/ACPLanguageModel | +| Source type | library docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the ACP search; says newer agents expose models as a `model` config option and older ones via an unstable models API. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/langchain-deepagents-acp-set-config-option.md b/research/cap-protocol/langchain-deepagents-acp-set-config-option.md new file mode 100644 index 0000000..68d2d49 --- /dev/null +++ b/research/cap-protocol/langchain-deepagents-acp-set-config-option.md @@ -0,0 +1,28 @@ +# deepagents-acp AgentServerACP.set_config_option + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | deepagents-acp AgentServerACP.set_config_option | +| Source URL | https://reference.langchain.com/python/deepagents-acp/server/AgentServerACP/set_config_option | +| Source type | library docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the ACP search; an ACP server implementation handling mode and model switches. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/lmsys-routellm-blog.md b/research/cap-protocol/lmsys-routellm-blog.md new file mode 100644 index 0000000..5479789 --- /dev/null +++ b/research/cap-protocol/lmsys-routellm-blog.md @@ -0,0 +1,50 @@ +# RouteLLM: An Open-Source Framework for Cost-Effective LLM Routing (LMSYS) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | RouteLLM: An Open-Source Framework for Cost-Effective LLM Routing (LMSYS) | +| Source URL | https://lmsys.org/blog/2024-07-01-routellm/ | +| Source publisher / author | LMSYS | +| Source type | blog post | +| Source date (published) | 2024-07-01 | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | Medium | +| Status | Active | + +## Summary + +The announcement of RouteLLM: routers that decide per query between one strong, expensive model and one weak, cheap model, trained on preference data. + +## Key findings + +- Setup is two models: a stronger, more expensive one and a weaker, cheaper one; routers are trained on preference data (Chatbot Arena, augmented with golden labels and an LLM judge). +- Reported cost reduction versus using only GPT-4: over 85% on MT Bench, 45% on MMLU, 35% on GSM8K, 'while still achieving 95% of GPT-4's performance'. +- On MT Bench, matrix factorization reaches 95% of GPT-4 quality with 26% GPT-4 calls; with LLM-judge augmentation 14% of calls. On MMLU the best causal LLM router needs 54% GPT-4 calls. +- Router types: similarity-weighted ranking, matrix factorization, BERT classifier, causal LLM classifier. +- Without augmentation the routers perform near random on MMLU (out-of-distribution); the cost model treats weak-model calls as negligible; the Claude/Llama generalization was tested on MT Bench only. + +## Pros / strengths + +Quantified cost and quality tradeoff for per-query routing between two tiers. + +## Cons / limitations / caveats + +Two-model setting on public benchmarks; savings depend on the workload; the post has no dedicated limitations section. Needs trained routers, a dependency tai avoids. + +## Relevance to tai + +Evidence that sending only the hard fraction to the strong model is worth doing, but via a trained router; for tai that argues for letting the calling model (or the user via `@`) decide, not for building a router. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/mcp-directory-sub-agents.md b/research/cap-protocol/mcp-directory-sub-agents.md new file mode 100644 index 0000000..98f1630 --- /dev/null +++ b/research/cap-protocol/mcp-directory-sub-agents.md @@ -0,0 +1,28 @@ +# sub-agents (mcp.directory) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | sub-agents (mcp.directory) | +| Source URL | https://mcp.directory/skills/sub-agents | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/medium-claude-code-subagent-model-routing.md b/research/cap-protocol/medium-claude-code-subagent-model-routing.md new file mode 100644 index 0000000..b824a7a --- /dev/null +++ b/research/cap-protocol/medium-claude-code-subagent-model-routing.md @@ -0,0 +1,28 @@ +# Claude Code subagent model routing + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Claude Code subagent model routing | +| Source URL | https://medium.com/@roanmonteiro/claude-code-subagent-model-routing-stop-paying-for-opus-on-haiku-work-ee76dc32cb88 | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search; recommends cheap models for read-only work. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/microsoft-learn-agent-framework-skills.md b/research/cap-protocol/microsoft-learn-agent-framework-skills.md new file mode 100644 index 0000000..afe75a3 --- /dev/null +++ b/research/cap-protocol/microsoft-learn-agent-framework-skills.md @@ -0,0 +1,28 @@ +# Agent Framework skills (Microsoft Learn) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent Framework skills (Microsoft Learn) | +| Source URL | https://learn.microsoft.com/en-us/agent-framework/agents/skills | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search; listing quoted a 1024-character description limit and a 500-line body guideline. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/mintlify-openfang-a2a.md b/research/cap-protocol/mintlify-openfang-a2a.md new file mode 100644 index 0000000..e9b65e3 --- /dev/null +++ b/research/cap-protocol/mintlify-openfang-a2a.md @@ -0,0 +1,28 @@ +# A2A integration (OpenFang) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | A2A integration (OpenFang) | +| Source URL | https://www.mintlify.com/RightNow-AI/openfang/integrations/a2a | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the A2A search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/modelcontextprotocol-sampling-model-preferences.md b/research/cap-protocol/modelcontextprotocol-sampling-model-preferences.md new file mode 100644 index 0000000..9611653 --- /dev/null +++ b/research/cap-protocol/modelcontextprotocol-sampling-model-preferences.md @@ -0,0 +1,50 @@ +# MCP specification 2025-06-18: Sampling (model preferences) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | MCP specification 2025-06-18: Sampling (model preferences) | +| Source URL | https://modelcontextprotocol.io/specification/2025-06-18/client/sampling | +| Source publisher / author | Model Context Protocol project | +| Source type | standard | +| Source date (published) | 2025-06-18 | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +The MCP feature by which a server asks the client's model for a completion, notable here for its model-selection abstraction. + +## Key findings + +- A server cannot simply request a model by name because the client may not have that model or may prefer another provider's equivalent. +- Preferences combine three normalized priorities, `costPriority`, `speedPriority`, `intelligencePriority` (0 to 1), with optional `hints`. +- Hints are substrings that match model names flexibly, are evaluated in order, MAY map to equivalent models from other providers, and are advisory: 'clients make final model selection'. +- Example: hints `claude-3-sonnet` then `claude`, with cost 0.3, speed 0.8, intelligence 0.5. +- Humans in the loop: there SHOULD always be a way to deny sampling requests; clients SHOULD show requests and responses for review, rate-limit, and respect hints. + +## Pros / strengths + +A finished, widely implemented answer to 'how does a caller name a model without knowing the callee's catalog': substring hints plus capability priorities, resolved by the owner of the models. + +## Cons / limitations / caveats + +Priorities are unitless hints; the spec does not say how a client turns them into a choice. + +## Relevance to tai + +Matches the shape of tai's fuzzy `@model` resolution (substring-style hints resolved by the side that knows the catalog). A tool input of `hints` plus optional priorities is a precedent for `ask_agent(model_hint=...)`. Human-in-the-loop rules inform the approval default. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/mohakchugh-llm-routing-cascade.md b/research/cap-protocol/mohakchugh-llm-routing-cascade.md new file mode 100644 index 0000000..2700f0a --- /dev/null +++ b/research/cap-protocol/mohakchugh-llm-routing-cascade.md @@ -0,0 +1,28 @@ +# LLM routing and cascade inference + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | LLM routing and cascade inference | +| Source URL | https://mohakchugh.is-a.dev/blog/llm-routing-cascade-inference-cost-quality-optimization | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search; unreferenced 60-80% small-model traffic claim. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/morphllm-a2a-protocol.md b/research/cap-protocol/morphllm-a2a-protocol.md new file mode 100644 index 0000000..cfe888d --- /dev/null +++ b/research/cap-protocol/morphllm-a2a-protocol.md @@ -0,0 +1,28 @@ +# Google's A2A Protocol: What It Actually Does + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Google's A2A Protocol: What It Actually Does | +| Source URL | https://www.morphllm.com/a2a-protocol | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the A2A search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/neuraltrust-llm-model-routing.md b/research/cap-protocol/neuraltrust-llm-model-routing.md new file mode 100644 index 0000000..a05c63e --- /dev/null +++ b/research/cap-protocol/neuraltrust-llm-model-routing.md @@ -0,0 +1,28 @@ +# What is LLM model routing? + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | What is LLM model routing? | +| Source URL | https://neuraltrust.ai/blog/llm-model-routing | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/openai-agents-python-multi-agent-orchestration.md b/research/cap-protocol/openai-agents-python-multi-agent-orchestration.md new file mode 100644 index 0000000..650eeab --- /dev/null +++ b/research/cap-protocol/openai-agents-python-multi-agent-orchestration.md @@ -0,0 +1,49 @@ +# OpenAI Agents SDK: Orchestrating multiple agents + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | OpenAI Agents SDK: Orchestrating multiple agents | +| Source URL | https://openai.github.io/openai-agents-python/multi_agent/ | +| Source publisher / author | OpenAI | +| Source type | vendor docs | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +Short orchestration guide separating LLM-driven from code-driven orchestration and agents-as-tools from handoffs. + +## Key findings + +- LLM-driven orchestration lets the model plan and choose steps; code-driven orchestration puts application code in charge of which agents run in what order, making speed, cost and performance more predictable. +- Agents as tools: use when a specialist handles a limited subtask 'but should not take over the user-facing conversation'; the manager keeps control of the reply. +- Handoffs: use when 'routing itself is part of the workflow'; the specialist owns the rest of the turn. +- Best practices: invest in good prompts, monitor and iterate, let the agent introspect and improve, build specialists that each excel at one task, use evals. + +## Pros / strengths + +Names the decision rule (who owns the reply) in two lines. + +## Cons / limitations / caveats + +Vendor guidance; no data. Overlaps with Anthropic's orchestrator-workers description. + +## Relevance to tai + +tai's `@agent prompt` is a handoff-like human-chosen route; a model-callable `ask_agent` would be the agent-as-tool form. This page backs choosing the tool form when tai must stay in control of the answer. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/openai-agents-python-tools-agent-as-tool.md b/research/cap-protocol/openai-agents-python-tools-agent-as-tool.md new file mode 100644 index 0000000..bb77c25 --- /dev/null +++ b/research/cap-protocol/openai-agents-python-tools-agent-as-tool.md @@ -0,0 +1,51 @@ +# OpenAI Agents SDK: Tools (agents as tools) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | OpenAI Agents SDK: Tools (agents as tools) | +| Source URL | https://openai.github.io/openai-agents-python/tools/ | +| Source publisher / author | OpenAI | +| Source type | vendor docs | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | High | +| Status | Active | + +## Summary + +The SDK page that defines `Agent.as_tool()`: exposing a specialist agent to a manager as a callable tool instead of handing the conversation over. + +## Key findings + +- The agent-as-tool run has its own state: the parent run's conversation state is not inherited automatically; pass the same `session` to share client-managed history. +- The nested run's final output is returned to the parent as the tool result; `custom_output_extractor` can reshape it first. +- `is_enabled` takes a boolean, a callable or an async function; disabled tools are completely hidden from the LLM at runtime. +- By default the tool expects an object with one string field `input`; a Pydantic model or dataclass via `parameters` replaces that, and `include_input_schema=True` adds the JSON Schema to the nested input. +- If `needs_approval` requires a decision, the nested run pauses and the pending items surface in `result.interruptions`. +- Guidance: use a central agent that orchestrates specialists 'instead of handing off control'. + +## Pros / strengths + +Concrete, small API that names the knobs a delegation tool needs: input schema, output extractor, enable switch, approval pause. + +## Cons / limitations / caveats + +Describes the SDK's API, not measured outcomes; the handoff comparison lives on another page. + +## Relevance to tai + +Template for the shape of a tai `ask_agent` tool: structured input, nested run in its own context, only the final output returned, an enable switch and an approval hook. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/openai-community-handoff-implementation.md b/research/cap-protocol/openai-community-handoff-implementation.md new file mode 100644 index 0000000..932acc0 --- /dev/null +++ b/research/cap-protocol/openai-community-handoff-implementation.md @@ -0,0 +1,28 @@ +# Openai-agents handoff implementation + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Openai-agents handoff implementation | +| Source URL | https://community.openai.com/t/openai-agents-handoff-implementation/1370797 | +| Source type | forum thread | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search; handoff versus tool discussion. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/openai-community-handoffs-cant-iterate.md b/research/cap-protocol/openai-community-handoffs-cant-iterate.md new file mode 100644 index 0000000..9797ba7 --- /dev/null +++ b/research/cap-protocol/openai-community-handoffs-cant-iterate.md @@ -0,0 +1,28 @@ +# OpenAI agents SDK - Handoffs can't iterate + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | OpenAI agents SDK - Handoffs can't iterate | +| Source URL | https://community.openai.com/t/openai-agents-sdk-handoffs-cant-iterate-am-i-right/1235072 | +| Source type | forum thread | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search; users hit the one-handoff-per-turn limit and move to agents as tools. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/openrouter-advisor-server-tool.md b/research/cap-protocol/openrouter-advisor-server-tool.md new file mode 100644 index 0000000..679fcae --- /dev/null +++ b/research/cap-protocol/openrouter-advisor-server-tool.md @@ -0,0 +1,50 @@ +# OpenRouter: Advisor server tool + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Dossier | +| Tier | 1 | +| Source title | OpenRouter: Advisor server tool | +| Source URL | https://openrouter.ai/blog/announcements/advisor-server-tool/ | +| Source publisher / author | OpenRouter | +| Source type | blog post | +| Source date (published) | Unknown — not stated | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | direct page fetch | +| Confidence | Medium | +| Status | Active | + +## Summary + +OpenRouter's cross-vendor advisor: any executor model can consult a stronger model mid-generation; the advice returns as a tool result and the executor writes the answer. + +## Key findings + +- The advisor is 'a consultant, not a ghostwriter': single-turn, no nested tool loop; a tool-carrying sub-agent is a separate tool. +- Parameters: `model`, optional `name`, `instructions`, `stream`; the executor passes a `prompt` describing what it needs help with. +- Safeguards: an `x-openrouter-advisor-depth` header on inner calls and stripping the advisor tool from sub-calls prevent unbounded nesting; consultations are capped per request (number not given). +- Advisor tokens bill at the advisor model's own rates, shown separately from the executor's. +- Claimed use: the 10-20% of requests with architectural decisions or ambiguous edge cases; in a 50-tool-call session the article estimates maybe 2-3 consultations. + +## Pros / strengths + +Names the safeguards a model-callable model needs: depth marker, tool stripping, per-request cap, separate billing. + +## Cons / limitations / caveats + +Vendor announcement; the cap value and the 10-20% share are asserted, not measured; a search listing suggested other pages describe it as a sub-agent, so docs may differ. + +## Relevance to tai + +Blueprint for tai's guard rails if a model-as-tool is added: nested calls get the tool removed, a static per-turn ceiling, and the cost line attributed to the consulted model. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Initial dossier created. | diff --git a/research/cap-protocol/originhq-escaping-sandbox-confused-deputies.md b/research/cap-protocol/originhq-escaping-sandbox-confused-deputies.md new file mode 100644 index 0000000..c12a1e3 --- /dev/null +++ b/research/cap-protocol/originhq-escaping-sandbox-confused-deputies.md @@ -0,0 +1,28 @@ +# Escaping the sandbox: confused deputies (Origin) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Escaping the sandbox: confused deputies (Origin) | +| Source URL | https://www.originhq.com/blog/escaping-the-sandbox-confused-deputies | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the delegation-security search; coding-agent research on injected prompts moving to less-sandboxed contexts. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/orq-ai-advisor-server-tool.md b/research/cap-protocol/orq-ai-advisor-server-tool.md new file mode 100644 index 0000000..d181246 --- /dev/null +++ b/research/cap-protocol/orq-ai-advisor-server-tool.md @@ -0,0 +1,28 @@ +# orq.ai advisor server tool + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | orq.ai advisor server tool | +| Source URL | https://docs.orq.ai/docs/ai-gateway/features/server-tools/advisor.md | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the advisor search; advisor receives the transcript and the model's question. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/patmcguinness-agent-architecture-debate.md b/research/cap-protocol/patmcguinness-agent-architecture-debate.md new file mode 100644 index 0000000..07f7f13 --- /dev/null +++ b/research/cap-protocol/patmcguinness-agent-architecture-debate.md @@ -0,0 +1,28 @@ +# The AI Agent Architecture Debate + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | The AI Agent Architecture Debate | +| Source URL | https://patmcguinness.substack.com/p/the-ai-agent-architecture-debate | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the multi-agent search; commentary on the Anthropic post. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/plainenglish-claude-code-spawning-agents.md b/research/cap-protocol/plainenglish-claude-code-spawning-agents.md new file mode 100644 index 0000000..6062caf --- /dev/null +++ b/research/cap-protocol/plainenglish-claude-code-spawning-agents.md @@ -0,0 +1,28 @@ +# Claude Code Has Been Spawning AI Agents Without Telling You + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Claude Code Has Been Spawning AI Agents Without Telling You | +| Source URL | https://ai.plainenglish.io/claude-code-has-been-spawning-ai-agents-without-telling-you-463db029e038 | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/resumelens-routing-vs-cascading.md b/research/cap-protocol/resumelens-routing-vs-cascading.md new file mode 100644 index 0000000..62bf538 --- /dev/null +++ b/research/cap-protocol/resumelens-routing-vs-cascading.md @@ -0,0 +1,28 @@ +# Two patterns: routing vs cascading + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Two patterns: routing vs cascading | +| Source URL | https://www.resumelens.org/blog/ai/llm-routing-cascades | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search; compares classifier routing and cascades, FrugalGPT described only secondhand. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/rubydoc-phone-a-friend.md b/research/cap-protocol/rubydoc-phone-a-friend.md new file mode 100644 index 0000000..89fec7b --- /dev/null +++ b/research/cap-protocol/rubydoc-phone-a-friend.md @@ -0,0 +1,28 @@ +# RubynCode PhoneAFriend tool + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | RubynCode PhoneAFriend tool | +| Source URL | https://www.rubydoc.info/gems/rubyn-code/0.9.0/RubynCode/Tools/PhoneAFriend | +| Source type | library docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the advisor search; a one-shot, tool-less consult of another provider's top model. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/scworld-mcp-confused-deputy.md b/research/cap-protocol/scworld-mcp-confused-deputy.md new file mode 100644 index 0000000..7baf0e9 --- /dev/null +++ b/research/cap-protocol/scworld-mcp-confused-deputy.md @@ -0,0 +1,28 @@ +# After the identity fix: MCP's confused deputy problem + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | After the identity fix: MCP's confused deputy problem | +| Source URL | https://www.scworld.com/perspective/after-the-identity-fix-mcps-confused-deputy-problem | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the delegation-security search; credentials are checked, intent is not. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/shekhargulati-routellm-paper.md b/research/cap-protocol/shekhargulati-routellm-paper.md new file mode 100644 index 0000000..c421e04 --- /dev/null +++ b/research/cap-protocol/shekhargulati-routellm-paper.md @@ -0,0 +1,28 @@ +# RouteLLM Paper (Shekhar Gulati) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | RouteLLM Paper (Shekhar Gulati) | +| Source URL | https://shekhargulati.com/2024/07/09/routellm-paper/ | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the routing search; a summary of the paper. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/sim-ai-agent-skills.md b/research/cap-protocol/sim-ai-agent-skills.md new file mode 100644 index 0000000..a9c8ad7 --- /dev/null +++ b/research/cap-protocol/sim-ai-agent-skills.md @@ -0,0 +1,28 @@ +# Agent skills (Sim docs) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent skills (Sim docs) | +| Source URL | https://docs.sim.ai/agents/skills | +| Source type | vendor docs | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search; a platform's skill documentation. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/simonwillison-openai-agents-sdk.md b/research/cap-protocol/simonwillison-openai-agents-sdk.md new file mode 100644 index 0000000..319579d --- /dev/null +++ b/research/cap-protocol/simonwillison-openai-agents-sdk.md @@ -0,0 +1,28 @@ +# OpenAI Agents SDK (Simon Willison) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | OpenAI Agents SDK (Simon Willison) | +| Source URL | https://simonwillison.net/2025/Mar/11/openai-agents-sdk | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the agents-as-tools search; an early commentary on the SDK. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/skillscat-authoring-skills.md b/research/cap-protocol/skillscat-authoring-skills.md new file mode 100644 index 0000000..e2a974b --- /dev/null +++ b/research/cap-protocol/skillscat-authoring-skills.md @@ -0,0 +1,28 @@ +# authoring skills (skills.cat) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | authoring skills (skills.cat) | +| Source URL | https://skills.cat/skills/lidessen/moniro/authoring-skills | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/specification-website-a2a-agent-cards.md b/research/cap-protocol/specification-website-a2a-agent-cards.md new file mode 100644 index 0000000..b1a6e55 --- /dev/null +++ b/research/cap-protocol/specification-website-a2a-agent-cards.md @@ -0,0 +1,28 @@ +# A2A agent cards (specification.website) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | A2A agent cards (specification.website) | +| Source URL | https://www.specification.website/spec/agent-readiness/a2a-agent-cards/ | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the A2A search; lists required card fields. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/superdatascience-pt1012-transcript.md b/research/cap-protocol/superdatascience-pt1012-transcript.md new file mode 100644 index 0000000..82e0593 --- /dev/null +++ b/research/cap-protocol/superdatascience-pt1012-transcript.md @@ -0,0 +1,28 @@ +# SuperDataScience PT1012 transcript + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | SuperDataScience PT1012 transcript | +| Source URL | https://www.superdatascience.com/wp-content/uploads/PT1012-Transcript.pdf | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the advisor search; single source for Haiku about 20% versus about 41% with an Opus advisor on BrowseComp, unverified. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/tembo-claude-code-subagents-guide.md b/research/cap-protocol/tembo-claude-code-subagents-guide.md new file mode 100644 index 0000000..2c4fad6 --- /dev/null +++ b/research/cap-protocol/tembo-claude-code-subagents-guide.md @@ -0,0 +1,28 @@ +# Claude Code Subagents: A 2026 Practical Guide + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Claude Code Subagents: A 2026 Practical Guide | +| Source URL | https://tembo.io/blog/claude-code-subagents | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the subagent-model search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/tessl-aider-desk-skill-creator.md b/research/cap-protocol/tessl-aider-desk-skill-creator.md new file mode 100644 index 0000000..1343941 --- /dev/null +++ b/research/cap-protocol/tessl-aider-desk-skill-creator.md @@ -0,0 +1,28 @@ +# skill-creator (tessl registry) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | skill-creator (tessl registry) | +| Source URL | https://tessl.io/registry/skills/github/hotovo/aider-desk/skill-creator | +| Source type | other | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the skills search; listing gave token sizes for metadata and body. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/wandb-a2a-guide.md b/research/cap-protocol/wandb-a2a-guide.md new file mode 100644 index 0000000..f7f6f61 --- /dev/null +++ b/research/cap-protocol/wandb-a2a-guide.md @@ -0,0 +1,28 @@ +# Agent2Agent (A2A) guide (Weights and Biases) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Agent2Agent (A2A) guide (Weights and Biases) | +| Source URL | https://wandb.ai/byyoung3/Generative-AI/reports/Agent2Agent-A2A-A-guide-for-AI-integration--VmlldzoxMjQwMjkwNg | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the A2A search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. | diff --git a/research/cap-protocol/zenml-multi-agent-research-claude.md b/research/cap-protocol/zenml-multi-agent-research-claude.md new file mode 100644 index 0000000..9d5cd84 --- /dev/null +++ b/research/cap-protocol/zenml-multi-agent-research-claude.md @@ -0,0 +1,28 @@ +# Building production multi-agent research systems with Claude (ZenML) + +## Metadata + +| Field | Value | +|---|---| +| Institute | Molodetz Institute — Research & Retrieval Division | +| Subject | cap-protocol | +| Document type | Source Citation Card | +| Tier | 2 | +| Source title | Building production multi-agent research systems with Claude (ZenML) | +| Source URL | https://zenml.io/llmops-database/building-production-multi-agent-research-systems-with-claude | +| Source type | blog post | +| Date researched | 2026-10-25 | +| Researched by | tai (agent), model: Claude Sonnet 5, on behalf of retoor | +| Research method | Search aggregator result listing (not independently opened or quoted) | +| Confidence | Not independently verified | +| Status | Active | + +## Why it's here + +Surfaced by the multi-agent search. + +## Changelog + +| Date | Change | +|---|---| +| 2026-10-25 | Logged as consulted-but-not-analyzed. |