Tracker: `GOVERNANCE.md > Durable Knowledge and Self-Improvement`

[ptr727]

Tracker for GOVERNANCE.md > Durable Knowledge and Self-Improvement. Filed under #1315's roll-up of the open prose backlog: every finding a bundle issue enumerated against this unit is listed here once, one line each, naming the issue it came from, and the bundle is closed as a duplicate of this tracker so the finding is not lost and the bundle stops being a unit of work. A new pre-existing finding on this unit gathers here per pr-review-conduct's Merge Gate, numbering continuing this list. This tracker closes when the unit is rewritten and each line below is settled or superseded.

1. GOVERNANCE.md:22 repeats the same not-carried clause for agent-conduct's SKILL.md path. (Carried Text Says the Skills Are Not Carried, While the Manifest Carries Them to Every Repo #1184)
2. "Agent memory does not survive a new session" is false about the tooling: a per-machine memory file persists; what does not survive is a new machine. (Pre-Existing Defects the Canonical Review Surfaced in Units #1186 Touched #1196)
3. The three-condition gate is stated unconditionally then scoped to host hooks, and for a committed hook it relaxes, re-binds, and omits conditions inconsistently. (Pre-Existing Defects the Canonical Review Surfaced in Units #1186 Touched #1196)
4. "The involuntary layer that fires in every session" overstates a per-machine host hook install and is contradicted by this file's own note. (Pre-Existing Defects the Canonical Review Surfaced in Units #1186 Touched #1196)
5. "A chained Skill trigger" is a mechanism named nowhere else in the carried set. (Pre-Existing Defects the Canonical Review Surfaced in Units #1186 Touched #1196)

[ptr727]

Two further pre-existing findings from the whole-unit read the #1317 class 2 pull request ran over this unit, gathered here per the tracker's rule:

6. "A skill's own trigger going unread by the session at all, by contrast, is a loading problem, fixed by getting the rule into context (the CLAUDE.md importing AGENTS.md), not by a hook." The parenthetical names a fix that does not reach the failure it is attached to: the CLAUDE.md import loads AGENTS.md text, and a skill's trigger reaches a session only through the skills install, which AGENTS.md "Where the Rules Live" itself names as the fix for a rule that keeps needing restating.
7. "so it was promoted to a gh-write-guard hook rule." The unit locates its other two mechanisms (.husky/pre-push, the agent-conduct Skill path) but names gh-write-guard with no location, so a reader with only this unit cannot verify the promoted rule exists. It does, in host-setup/agent-safety/claude/gh-write-guard.py.

[ptr727]

Two more from the second whole-unit read on the same pull request:

8. "The triggering shape is decidable from the tool call's own text, arguments, and working directory alone" is the condition the bullet later says "still binds" a committed hook, but a committed git hook sees no tool call. It sees a git operation's refs, object ids, and tree. Applied literally to the .husky/pre-push example the unit itself gives, the condition does not fit its own wording.
9. "belongs in a committed governance file (GOVERNANCE.md for a cross-cutting rule, AGENTS.md, CODESTYLE.md, WORKFLOW.md, or a committed backlog the repository already keeps)" lists a backlog as a member of the set "governance file", which it is not, so the parenthetical misstates its own category.