[Deep-Sixed]

Fixes from a second doc review of README.md and docs/agent-guide.md. Docs only; no code changes.

Operational gaps

• Missing upgrade note for 006_ref_updates.sql. Hubs set up before it reject every push until quorumgit hook install is re-run to add the reference-transaction hook. The README's upgrade notes, previously scattered, are now a single Upgrading section in migration order covering 004 to 011, with an explanation of why there is no 003 (that number was used only on branches that were never merged). The 010 and 011 notes that arrived from main in Compose ref policies and preserve quorum executor separation #35 are included in the same list.
• Recovering hooks after reinstalling QuorumGit. The hooks hard-code the path of the Python interpreter that ran hook install. If that interpreter moves (a Python upgrade, a rebuilt venv, a reinstall), pushes fail, and hook install refuses to overwrite its own now-different hooks ("Existing QuorumGit-managed hook differs"). The fix is documented: delete both hook files, then reinstall. I reproduced both the refusal and the recovery against a real bare hub, before and again after merging main.

Agent guide

• Step 6 told agents to release finished work, which returns the task to open. It now distinguishes three cases: task done for finished work, handoff when someone should continue, and release when stepping away from unfinished work.

Accuracy

• "Eight concepts" is now "Nine", matching the table.
• status is described as it behaves: it labels the path uri:, and prints running: false and exits 1 when no store exists. Verified by running it.
• task done is added to the list of audited actions.
• The command table notes that Git itself calls hook pre-receive and hook reference-transaction, so users never run them.

Merge

main (#32, #34, #35) is merged in. The one conflict was in the README upgrade notes, resolved by folding main's 010 and 011 notes into the consolidated section.

After the merge: ruff check is clean, pyright reports 0 errors, and pytest tests gives 283 passed.

🤖 Generated with Claude Code

https://claude.ai/code/session_01SHGwY6PLGwr7kQY63qQGq6