ci: a numbered idea pile and the Idea-Id trailer gate (E3-piles, E3-trailers)

[rudi193-cmd]

Bites: E3-piles-forge-workshop (docs: commit) and E3-trailers-forge-workshop (ci: commit). Fleet plan Wave 3; orchestrator session session_01Qa81BaErcUeJgGfBGRmNqm.

What

• docs/ideas.md — the numbered pile, in willow-reconciler's form. Three items: the first bundle cut (carried from the README's "After the first bite"), adopting Idea-Id trailers, and the Nestor pin raise for Wave 6. Nothing is hand-tagged: git history showed none of them landed when the doc was written.
• .github/workflows/trailers.yml — runs reconciler verify on push and PR. Branch list mirrors tests.yml (master, main) since this is a template and an instantiated workshop may default to either. permissions: contents: read added, matching tests.yml.
• CONTRIBUTING.md — created in the tree (the org's copy never reaches a clone, per CLAUDE.md). Names the test command and carries the "The Idea-Id commit-trailer convention" section.
• tests/test_release_wiring.py — stdlib-only structural check: trailers.yml exists wherever docs/ideas.md exists, plus the plant asserting the pile is present so the check is never vacuous.
• README.md — four rows added to the "What is here" table for the files above; the tests/ row no longer says one test ships.

Validate summary

Of 3 item(s) parsed from docs/ideas.md (./): 0 carry the doc's own explicit legend tag (0 landed, 0 partial — read back verbatim, not classified). Of the remaining 3 untagged item(s), 1 resolved to a real inferred signal (1 landed, 0 partial) and 2 show no reliable automatic landing signal.
- items parsed: 3 of 3 (0 dropped, unparseable)
- hold-out score: 0/0 recovered (recovery_rate=None)

reconciler verify --repo ./ --doc docs/ideas.md:

1 Idea-Id trailer(s) in ./'s history: 1 resolve to an item in docs/ideas.md (covering 1 distinct item(s)), 0 name an id the doc does not contain.
- ok `willow-ideas-002` (landed) in 9643e66473: ci: the Idea-Id trailer gate (E3-trailers)

Trailer lines the commits carry

• f890bba docs: — first commit, no Idea-Id (the pile did not exist before it).
• 9643e66 ci: — Idea-Id: willow-ideas-002, generated by reconciler id --repo ./ --doc docs/ideas.md --grep "Idea-Id".

Verified

• python -m pytest tests/ -q → 4 passed (the two existing tests plus the two new ones).
• python -m unittest tests.test_release_wiring → OK (stdlib runner, no pytest).
• reconciler run --repo ./ --doc docs/ideas.md --validate → exit 0, 0 dropped.
• reconciler verify --repo ./ --doc docs/ideas.md → exit 0, 1 resolved, 0 unresolved.
• trailers.yml parses as YAML. No .db file is tracked.

--repo ./, not --repo .

willow-reconciler 0.6.0 treats --repo as a path only when it carries a slash or is absolute; a bare . is resolved as a repo name beside the checkout and errors out (repo '.' not found). Every command written in this PR therefore says --repo ./, with a comment saying why in the workflow and in CONTRIBUTING. The fix belongs upstream (treat . and .. as path-like in cli._is_path_like), or the fleet text should say ./. Outside this bite; flagged for the orchestrator.

Seen, not touched

• No release-please here. This repo has no release-please.yml, no release-please-config.json, and no auto-merge, so no pr-title guard is required today. If Wave 4 adds release-please to this repo, the guard must land with it.
• CLAUDE.md still says the org keeps CONTRIBUTING.md in forge-play/.github. True, but a copy now also lives here. Left alone: one sentence of context, not a rule.
• pyproject.toml pins nestor-meaning>=0.19.0,<1.0.0. Wave 6 raises it; item 3 of the pile records that.
• No tests/test_fleet_conventions.py exists in this repo; the reconciler's conventions verb describes one. Not part of this bite.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Qa81BaErcUeJgGfBGRmNqm

---

Generated by Claude Code