reports/ has two contradictory contracts: post-merge-systemize says never commit it, triage-friction-log says git-track it

[topij]

Two shipped skills give opposite contracts for the same directory, and .gitignore matches neither.

The contradiction

.claude/commands/post-merge-systemize.md:286 — commit only the paths you edited:

commit only the doc/skill/config paths you changed — never reports/, state/, or data files

.claude/commands/triage-friction-log.md:442:

Delete the state file. The reports/triage_<today>.md report is kept (git-tracked alongside other reports, no automatic cleanup).

triage-friction-log.md:184 reinforces the second reading, telling the operator the report is "also pushed in next branch".

Why it bites

.gitignore ignores state/ but not reports/. So post-merge-systemize's rule is unenforced: an operator who follows it correctly is left with a permanently untracked reports/ in every git status, forever. Observed during the 2026-07-27 triage sweep (PR #78) — ?? reports/ showed up in every status check of the session, and the report was deliberately held out of the commit rather than silently introduce a reports/ convention into a repo that ships to adopters.

Which way it should resolve

Toward ignored, for four reasons beyond the inconsistency:

1. Derived. After a triage sweep lands, every proposal is an issue body and every source entry is verbatim in the friction-log archive. Nothing in the report is unrecoverable.
2. Unbounded with no budget. One file per run plus "no automatic cleanup". Every other narrative file in the kit has a doc_budgets tripwire; a tracked reports/ would be the sole exception — the kit violating Principle docs: step-by-step parallel-sessions how-to #1 by design.
3. state/ is already ignored on identical logic, including init.sh:731 appending it to adopters. The report is scratch from the same pipeline run.
4. This repo ships to adopters. A tracked reports/ becomes inherited structure nobody asked for — the same kit-owned/adopter-owned boundary problem as The cp -r quickstart can't distinguish kit-owned from adopter-owned files #18.

The honest counter-argument, recorded so it is not lost: the report is the only artifact showing what the LLM proposed versus what was filed. On a run where the operator skips or rewrites items, that delta is genuine provenance. But a git-tracked draft is a weak place to hold it — nobody reads old drafts. The better home is the graduation marker, which already summarizes outcomes and does get read.

Proposed change

• add reports/ to .gitignore, and to init.sh's add_ignore_line block so adopters get it too;
• amend triage-friction-log.md:442 (and the "also pushed in next branch" line at :184) to match post-merge-systemize;
• consider recording proposed-vs-filed counts in the graduation marker, where the provenance would actually be read.

Related

• The cp -r quickstart can't distinguish kit-owned from adopter-owned files #18 — kit-owned vs adopter-owned boundary
• Extract the friction-triage engine from cs-toolkit behind a tracker adapter #6 — the triage engine is not vendored yet, so the reports/ convention has never actually been exercised by a script. Cheapest possible moment to settle it.