fix: point workflows at the #346 review-mode defect fixes, prune closed triage

[m2ux]

The main-repo half of the #346 review-mode definition fixes. The definitions themselves are in #351, which targets workflows.

Merge #351 first. This PR's submodule pointer resolves only once that branch's commit is on the remote's workflows branch.

What is here

Submodule pointer → the #351 commit.

Guard triage prune (scripts/binding-fidelity-triage.json, 4 entries). Each recorded an orphan-input finding for a value work-package read with no producer: assumption_decisions, issue_request, and comprehension_dir on two techniques. DI-20 closed all four by declaring them as workflow variables, so the entries now match nothing — and a stale entry is reported exactly like an untriaged finding, so leaving them turns binding-fidelity red.

Walk re-baseline (6 snapshots + corpus stamp), per the procedure the stamp mismatch prints. The path diff is the intended step change and nothing else:

• every policy loses the retrospective-confirm checkpoint and gains announce-retrospective and verify-artifact-links. Worth noting what the old snapshot recorded for that checkpoint: setVariable: undefined on its only reachable option — decisionless ceremony on the face of it, which is why DI-19 converts it to a message rather than keeping the gate.
• the review-mode policy additionally reaches create-complete-doc, render-token-usage and publish-close-out-artifacts — the three steps DI-18, DI-21 and DI-32 open on the review path.

Why one PR and not three

The prune, the pointer and the baselines are one matched set. Apart, any half turns a guard or the suite red: the prune without the declarations reports four untriaged findings, the declarations without the prune reports four stale entries, and the pointer without the re-baseline fails six snapshots plus the stamp check.

Verification

• 17/17 guards pass against the new corpus.
• 742 tests pass. One failure, session-crypto's EACCES case, is pre-existing and environmental — it fails identically with this change stashed, because the sandbox this ran in presents / such that mkdir returns ENOENT where the test asserts EACCES.
• The corpus stamp and the submodule pointer both name the same commit.

Refs #346

🤖 Generated with Claude Code