Files
msd-core/.out-of-scope/commit-gate-reuse-capability-loop-point.md
Tom Boucher b65d04c044 docs(#3353): record #3346 and #3353 triage decisions as out-of-scope (#3362)
* docs(#3346): record new-host-as-in-tree-runtime as out-of-scope

New host runtimes go out-of-tree as EoS host-plugins listed in the EoS
Registry (ADR-1239), not as first-party in-tree registry entries. Sibling
to omp-runtime-in-core.md; on-point precedent #2170 (Devin CLI).

* docs(#3353): record commit_gates config as out-of-scope

A parallel commit_gates config would duplicate the existing capability
gate system (command-exit-zero, #2008/ADR-2008). The route is a commit:pre
loop-point reusing the existing gate. Companion defect #3352 stays open.

---------

Co-authored-by: sim <sim@local>
2026-08-11 17:20:56 -04:00

5.7 KiB

A parallel commit_gates config for artifact preconditions in cmdCommit

Source: #3353 Decision: wontfix — closed as filed; the mechanism already exists in the capability gate system Date: 2026-08-11

Proposal summary

#3353 asked for a declarative pre-commit gate evaluated inside cmdCommit (src/commands.cts): a commit_gates array in .planning/config.json that matches files in the commit's scope, runs a shell command against each match, blocks the commit on non-zero (before staging, so a refusal leaves the index untouched), and is overridable via a CLI flag. The motivation was concrete and real — a workflow can assert "do not commit while X holds" about an artifact (e.g. an ungrounded REVIEWS.md), be entirely right, and watch the artifact get committed anyway because the check lives only in the prompt and a model can report a stale or skipped verdict as fresh.

This entry denies the commit_gates config as filed. It does not deny the underlying need.

Why GSD does not own this (as filed)

  • The exact mechanism already exists in the capability system. Since #2008 / ADR-2008, a capability can declare a gates entry with check.predicate.kind: "command-exit-zero" — "run your command, block the loop on non-zero" — at a chosen loop point, with blocking and onError semantics and a when config-key gate. See docs/how-to/command-exit-zero-gate.md and docs/reference/gate-predicates.md. The proposed commit_gates shape (command + block_on: "nonzero" + match + override) is the same mechanism in a different config.
  • A second config would fragment the architecture. commit_gates in .planning/config.json, fired from cmdCommit, would coexist with capability gates in capability.json, fired from the loop-resolver — two config locations, two override paths, two evaluators, and ambiguity about which fires when. GSD deliberately owns one gate system; the capability gate is it.
  • The right route is an extension of the existing system, not a parallel one. There is no commit:pre loop point today. The 12 canonical loop points (src/loop-resolver.cts CANONICAL_POINTS: the :pre/:post pairs for discuss, plan, verify, and ship, plus the four execute points — execute:pre, execute:wave:pre, execute:wave:post, execute:post) do not include commit. The loop-point vocabulary is closed but additive-only (docs/reference/capability-manifest.md), so adding a commit:pre point — and wiring cmdCommit to invoke the loop-resolver there — lets any capability declare the existing command-exit-zero gate at commit time. That extends the system you have rather than building a second one beside it.
  • EoS (ADR-1239) is unaffected either way. cmdCommit is engine-internal; a pre-commit gate touches none of the six host-integration interface points (command/dispatch/model/hooks/state/ artifact are about host↔engine, not git internals). This is recorded only to head off the misread that this decision is an EoS constraint — it is a capability-architecture constraint.

What this does NOT cover

This entry denies a commit_gates config separate from the capability gate system. It does not deny, and must never be cited against:

  • Adding a commit:pre (and/or commit:post) loop extension point so capabilities can declare command-exit-zero gates at commit time. This is the route the filing should take, and it is welcome as a capability-system proposal (new loop point + ADR-level schema note).
  • The companion defect. #3352 — reviewer evidence is never verified, raw per-lane output is rm -rf'd, and an ungrounded REVIEWS.md feeds /gsd-plan-phase --reviews — is confirmed-bug on its own merits and stays open. That a full fix will likely use a commit-point gate does not make the commit_gates config the right home for the gate mechanism.
  • New check.predicate kinds added to the existing capability gate evaluator — a separate, valid extension path.
  • Hardening cmdCommit's existing git-level checks (gitignore, branching strategy, staging failures) — those are unrelated to declarative artifact preconditions.

Re-open criteria

  • A need is shown for a commit-time gate that the capability loop-point model genuinely cannot express — e.g. a precondition that must fire when no capability is installed and that cannot itself be a loop point. This is narrow: the commit:pre point + command-exit-zero predicate covers the stated case (run a checker on a matched artifact, block on non-zero), so re-opening requires the loop-point route to have been tried and found structurally insufficient.