Enforce the no-AI-attribution commit policy · Issue #5 · cameron.stream/misaligned

Finding #

The repository says “No emoji and no AI attribution in commits” (AGENT.md:107-108), but nothing enforces that rule:

.githooks/ contains only pre-commit, which runs the corpus gate and cannot inspect a pending commit message.
None of the four Tangled workflows or tools/check.sh checks commit messages for attribution trailers/boilerplate.
Main commit 6c7bcf7242240c954e8ce53891aa7a9511cbd329 (2026-07-10, “Rename the Knowledge token family to Thought”) contains both the 👾 Generated with [Letta Code] boilerplate and Co-Authored-By: Letta Code <noreply@letta.com>.

There are older historical violations too, but the current-day main violation proves this is still an active workflow hole rather than only legacy history.

Expected #

Prevent new violations mechanically. Add a commit-msg hook (wired by the repository’s worktree tooling) that rejects the known generated-attribution footer, AI co-author trailers, and emoji under the documented policy. Mirror the check server-side over the incoming commit range so bypassed/missing local hooks cannot land violations.

Add shell fixtures for accepted human commit bodies and rejected attribution/emoji cases. Do not rewrite published history as part of this issue; the goal is stopping recurrence.

Resolution #

Cameron rescinded the repository-specific ban on AI attribution on 2026-07-10. Commit 7c9bc49 updates AGENT.md and wiki/process/workflows.md to allow accurate AI contributor provenance while retaining descriptive commit prose and explicit staging. Enforcement is therefore neither needed nor desired.