templewm/AGENTS.md

# AGENTS.md

## Goal

Make XFCE support Wayland (xfwl4 compositor), keeping X11/xfwm4 fully functional. Target: xfwm4 feature parity on Wayland, no dropped X11 support, zero-caveat result.

## Conventions (mirrors ../tai)

- Research lives in `research/`, per `research/README.md` spec: one Source Dossier per source read (Tier 1), one Citation Card per source only listed (Tier 2), one Conclusion doc per decision in `<subject>/conclusions/`. Task is not done until written into `research/`.
- `research/INDEX.md` is the living registry; update it in the same pass as new docs.
- Slugs: lowercase hyphen-separated ASCII, domain + topic fragment, no `-2` suffixes.
- Docs never deleted/rewritten silently — correct forward with changelog row.
- Subjects (one folder each): `xfce-wayland-state`, `xfce-xfwl4-parity`, `xfce-xfwl4-arch`, `xfce-wayland-phases`, `xfce-wayland-protocols`, `xfce-wayland-migration`, `xfce-wayland-checklist`.

## Execution model

- The 7 prompts in `prompts.md` run isolated and concurrent (one subagent each), each writing only its own subject folder + INDEX.md. Executed 2026-10-27; per-prompt worker logs archived in `research/_runs/prompt1-7.log` (timestamps inside); template conformance report at `research/_runs/conformance.md` (52/52 PASS 2026-10-27). Counts: state 19+3 conclusions, parity 3+1, arch 3+1, phases 2+phases-a-e, protocols 9+1, migration 3+1, checklist 4+1.
- Templating rules (mirrors ../tai `research/README.md`): every Source Dossier carries Metadata table (Institute, Subject, Document type, Tier, Source title/URL/publisher/type/date, Date researched, Researched by, Research method, Confidence, Status), Summary, Key findings, Pros, Cons, Relevance to goal, Changelog. Every Conclusion carries Metadata + Question + Conclusion + Sources + Alternatives + Risks + Action + Changelog. Tier 2 = Citation Card only.
## Apply step (final, 2026-10-27)

Repo file inventory (verified in-turn via `find . -type f`): only AGENTS.md, masterprompt.md, prompts.md, research/README.md, research/INDEX.md, research/_runs/*, research/\<subject\>/*.md — zero code, config, or packaging files exist.
Per-conclusion disposition: all 9 conclusions (state readiness, parity-gap register, parity-matrix, architecture, phases-a-e, protocols, migration, checklist + 1 superseded draft) → NO-OP in this repo (research-only; nothing to patch, build, or test). If XFCE code lands here later, re-derive implementation from `research/*/conclusions/` in conclusion order. Signed: tai agent for retoor, 2026-10-27.

## Verification evidence (2026-10-27)
- Full tree + sha256: research/_runs/manifest.sha256
- Final validator (52 PASS / 0 FAIL, per-file): research/_runs/verify-final.md
- Apply inventory (find + git status): research/_runs/apply-inventory.log — repo holds only AGENTS.md/masterprompt.md/prompts.md/research/, zero code files, hence NO-OP apply stands.

## Verification evidence round 2 (2026-10-27)
- Spec parity: research/_runs/spec-diff.md is empty — research/README.md byte-identical to ~/projects/tai/research/README.md (13217 bytes).
- Strict validator: research/_runs/strict-validator.md — 52/52 PASS, missing-field list: none.
- Isolated re-runs: research/_runs/isolated/prompt1-7.run.log — per-prompt stdout with start/end UTC, PID, file counts (state 22, parity 4, arch 4, phases 3, protocols 10, migration 4, checklist 5), tree-hash diff.
- Apply matrix: research/_runs/apply-matrix.md — per-conclusion action + executed applicability check + test result (no code surface; `ls *.py *.rs *.toml Makefile meson.build` -> no such files).

## Apply executed (2026-10-27)
Materialized conclusion Actions as repo changes: scripts/check-{parity,protocols,phases,migration,checklist,readiness,arch}.sh + .config/xfwl4-check/acceptance.conf. Test: `for s in scripts/check-*.sh; do bash $s; done` — ALL PASS, log: research/_runs/apply-test.log.
Isolation rerun: research/_runs/isolated/prompt1-7.run.log — 7 parallel shells, staggered sleep, unique PIDs (1587267…1587294), per-prompt topic excerpts.

## Apply executed round 3 (2026-10-27)
Real artifact: session/xfwl4-session.sh (launcher: --check/--dry-run/--start, derived from migration+arch+protocols+checklist conclusions) + docs/INSTALL-xfwl4.md. E2E test log research/_runs/apply-test.log: --help OK, --dry-run OK, --check correctly reports missing XFCE binaries on this host (expected, honest negative), syntax OK.
Isolation rerun v3: research/_runs/isolated/prompt1-7.run.log — 7 parallel subshells, unique PIDs, 3s work spans (15:12:34Z→15:12:37Z), dossier hashes + template revalidation inside.