[instructions] Sync instruction files with release v0.89.21

[github-actions]

Summary

Instructions Janitor sync for .github/aw/, audited since release v0.89.21.

Release and change audit

Compared commits touching docs/ and .github/aw/ since v0.89.21 (2026-09-23) against the current instruction files, accounting for the prior janitor syncs already landed in that window (PR 63142, PR 63399, PR 63976, PR 64228). Found accuracy gaps from three feature PRs that landed after the last sync and never touched .github/aw/:

• PR 64174 — Add required-labels gating to assign-to-agent
• PR 64179 — Allow workflows to narrow add-labels item schemas
• PR 64593 — Support AWF task-level model routing in the Copilot engine

Investigated other post-sync changes (sandbox kebab-case field rename PR 64302, self-hosted runner compile-time enforcement PR 64173, runtimes.node override fix PR 64498, private-to-public-flows hardening PR 64267, the repo-memory ledger feature set) and confirmed no further .github/aw/ changes were needed: the sandbox rename only affects user-facing frontmatter examples that none of the instruction files reproduce, the runner-enforcement and runtimes.node fix are compile/governance-level behavior not covered by workflow-authoring instructions, PR 64267 already synced syntax-tools-imports.md inline, and the ledger feature set was already fully written up in memory.md / memory-stateful-patterns.md by PR 64599.

Size audit

wc -l .github/aw/*.md — all 70+ files remain under their target limits both before and after this change.

Duplication audit

Found one real duplication: memory-stateful-patterns.md's "Append-only Event History" section restated the ledger's shard/record/query limits, compaction behavior, and Cloud Hypervisor isolation model nearly verbatim from memory.md's "Structured event history" section. Trimmed the worked-example file down to the usage pattern and added a cross-link to memory.md for the authoritative reference, matching how the file already defers to memory.md elsewhere.

Accuracy audit

Updated against current schema/code (main_workflow_schema.json, assign_to_agent.go, model_routing.go):

• safe-outputs-automation.md — documented required-labels on assign-to-agent
• safe-outputs-management.md — documented item-schema on add-labels
• syntax-engine.md — documented engine.model-routing (Copilot-only, requires AWF firewall + minimum AWF version)

Files changed

File	Before	After
memory-stateful-patterns.md	157	142
safe-outputs-automation.md	306	307
safe-outputs-management.md	317	318
syntax-engine.md	92	93

All edited files remain under their .github/aw/*.md target limit (< 400 lines).

Test plan

• wc -l .github/aw/*.md confirms all files, including the four edited, stay under the 400-line target.
• Manually cross-checked each new documentation line against the corresponding Go source (assign_to_agent.go, add_labels_item_schema.go, model_routing.go) and JSON schema (main_workflow_schema.json) for field names, defaults, and constraints.
• No compiled workflow (.lock.yml) changes required — this PR only touches prose instruction files.

🤖 Generated with [Claude Code]((claude.com/redacted)

Generated by 🧹 Instructions Janitor · claude · sonnet50 · 375.3 AIC · ⌖ 29.3 AIC · ⊞ 6.4K · ◷

• expires on Oct 3, 2026, 12:54 AM UTC-08:00

---

Generated by 👨‍🍳 PR Sous Chef · pi · gpt54 · 33.8 AIC · ⌖ 8.59 AIC · ⊞ 9.7K · ◷
_{Comment /souschef to run again}

[gh-aw-bot]

pr-sous-chef
@copilot this PR still needs forward progress before a maintainer can investigate efficiently.

Please refresh the branch if needed, then run the pr-finisher skill.

A branch refresh was also requested by PR Sous Chef in this run.

Generated by 👨🍳 PR Sous Chef

Generated by 👨‍🍳 PR Sous Chef · pi · gpt54 · 33.8 AIC · ⌖ 8.59 AIC · ⊞ 9.7K · ◷
_{Comment /souschef to run again}

[Copilot]

pr-sous-chef
@copilot this PR still needs forward progress before a maintainer can investigate efficiently.
...

The branch was already up to date with main. I ran make fmt, make lint, make test-unit, make test, and make recompile; all passed, and the worktree remains clean, so no new commit was needed. No unresolved review threads were present. Existing PR checks are still action_required, and the Copilot check is in progress.

[gh-aw-bot]

@copilot please revisit this update, make sure any reviewer concerns are handled, refresh the branch if needed, and run the pr-finisher skill before handing it back.

Generated by 👨‍🍳 PR Sous Chef · pi · gpt54 · 25.3 AIC · ⌖ 8.76 AIC · ⊞ 9.4K · ◷
_{Comment /souschef to run again}

[Copilot]

@copilot please revisit this update, make sure any reviewer concerns are handled, refresh the branch if needed, and run the pr-finisher skill before handing it back.

Generated by [👨‍🍳 PR ...

The branch remains up to date with main, and no unresolved review threads are present. I reran make fmt, make lint, make test-unit, make test, and make recompile; all passed, and the worktree is clean, so no new commit was needed. The prior Copilot check succeeded; the latest is still in progress. The three earlier workflow runs are action_required with no jobs/logs, and GitHub still reports the PR as blocked.

[github-actions]

🎉 This pull request is included in a new release.

Release: v0.90.3