guardrails: block-noncanonical-commit file-based alias resolution misses alias.<name>.command subkey (git config --get returns empty)

[kyle-sexton]

Found during the PR #903 .command subkey fix round (adjacent, pre-existing, left out of that surgical scope).

block-noncanonical-commit.sh:261 resolves file-based aliases via git config --get alias.$sub, which returns empty for a file-based alias.<name>.command definition — yet git (verified 2.54.0.windows.1) honors the .command subkey as an alias. A commit-bypassing alias written to config via the subkey spelling is therefore not resolved and not checked.

Class: under-detection (fail-open) in a security guard, BUT the threat model requires a prior git-config WRITE (config-file mutation), unlike the command-line -c/--config-env injection the #903 fix closed — narrower exposure, config-write paths have their own guard surface.

Fix direction: resolve both spellings — git config --get alias.$sub OR git config --get alias.$sub.command (last-wins semantics across the two spellings match git's behavior, verified empirically in the #903 fix round; see the cross-form last-wins tests added to lib/hook-utils.test.sh).

Origin: fixer round on PR #903, 2026-07-22.

[kyle-sexton]

This was generated by AI (triage lane — gate-audit claim). Claiming this item for one gate-audit cycle: an independent panel re-classifies whether the needs-human / status: needs-decision gate is still warranted, or whether the item is C2/C3 and liftable to agent-ready. Outcome comment follows in this cycle.

[kyle-sexton]

This was generated by AI (triage lane — gate audit). Independent panel: two fresh-context agents (one fable, one opus), each given only the issue content and the repo's own plugins/autonomy/reference/guardrails/work-classes.md, plus the triage lane's own read. This lane never executes work items, authors fixes, or merges.

Gate audit: GATE UPHELD — C3 shape, security-surface floor

Panel: 2-of-2 STAYS-GATED. Lane read concurs.

The fix is tightly specified — resolve both alias.$sub and alias.$sub.command with git's last-wins semantics, with an existing test pattern in lib/hook-utils.test.sh to mirror. But it modifies bypass-detection logic inside a commit guard, where an implementation slip silently widens a fail-open. That is the security floor, and excellent briefing does not clear it.

Stale pointer — important for whoever implements this: one leg verified the issue's cited location has moved. block-noncanonical-commit.sh:261 is now block-convention-violation.sh:310, still resolving only git config --get "alias.$sub". The sibling guard block-dangerous-git.sh already re-checks both spellings, so the correct behaviour is demonstrated in-repo.

Decision required

Ratify the near-mechanical guard change as scoped, or take the larger shared-helper refactor?

RECOMMENDED — approve as scoped, route through human merge. Add alias.$sub.command resolution with last-wins semantics at block-convention-violation.sh:310, mirroring the sibling guard that already does it right, with cross-form contract tests following the #903 round. Smallest diff that closes a live fail-open, and the correct implementation already exists next door to copy.

Alternative — extract the dual-spelling resolution into lib/hook-utils.sh so both guards share one implementation, eliminating the drift that caused this. Better end state, but it is a shared-lib change that fans out to the synced per-plugin copies — larger blast radius, and worth doing as a deliberate follow-up rather than folded into a fail-open fix.

[kyle-sexton]

This was generated by AI (triage lane — claim release). Gate-audit cycle complete for this item; the triage claim posted earlier this cycle is released. No lease is held — another instance or lane may pick this up immediately. The outcome comment above this one carries the classification and either the lift rationale or the decision summary.