Skip to content

Phase 2: Enforce /squad review as a required check and gate epic relay on verdict #1734

Open

Description

@bradygaster

Parent: #1729
Depends on: #1733

Goal

Promote /squad review from advisory to enforcing, once Phase 1 has produced enough real review output to judge quality.

Acceptance criteria

  • Set inlined-imports: true and pin a stable job name so a repository ruleset can require the check.
  • Required status check Squad Review / review enabled on dev first, then main after a soak period.
  • Epic merge relay in workflows/squad-implement-worker.md refuses to advance to the next child when the merged PR carried an unresolved REQUEST_CHANGES verdict from the reviewer.
  • A documented, auditable override path for the case where the reviewer is wrong and a human needs to proceed anyway.

Notes

Two ordering constraints:

  • Ruleset checks need inlined-imports: true to resolve reliably, so that flag change belongs here rather than in Phase 1.
  • Gating the epic relay is the change that actually makes review part of the lifecycle rather than a bolt-on. Without it, review can request changes while the epic marches on to the next child.

Do not enable the required check until Phase 1 has been running advisory long enough to show the reviewer is not producing false REQUEST_CHANGES verdicts at a rate that would stall the pipeline.

Activity

  1. bradygaster commented on Aug 23, 2026

    @bradygaster
    CollaboratorAuthor

    Procedures triage brief — Phase 2 enforcement

    Owner & rationale

    Owner: Booster (squad:booster). The risky change is turning review into a required status check/ruleset gate and making generated workflow output lintable. FIDO should co-review verdict semantics, but Booster owns CI/CD enforcement.

    Files in scope

    Verified existing paths:

    • workflows\squad-implement-worker.md
    • est\gh-aw-quality.test.ts
    • est\gh-aw-implement-workflow.test.ts
    • .github\workflows\squad-ci.yml
    • .github\workflows\squad-workflow-lint.yml

    Phase input created by #1733: workflows\squad-review.md.

    Prompt-budget impact

    No expected workflows\squad.md growth: +0 KB to the router prompt. If enforcement needs router override text, treat that as a design change against the remaining ~35.6–37.6 KB headroom and budget-review it before implementation.

    Dependencies

    Verification bar

    Seed a PR/review fixture with an unresolved REQUEST_CHANGES verdict and prove the epic relay refuses to advance; then resolve/override it and prove the relay can advance. A no-op implementation that only declares a job name must fail. Generated .lock.yml must pass actionlint or the gate is not trustworthy.

    Out of scope

    Building the advisory reviewer itself, remediation workers, and docs.

    Session-stall guard

    Before starting: this repo has phantom CRLF diffs (#1793). If git status shows files modified that you did not touch, run git diff --ignore-cr-at-eol. If that output is empty, the diff is line-ending churn, not real edits — run git add --renormalize . and continue. Do not halt and ask for confirmation; halting on phantom diffs is the documented cause of stuck sessions in this repo.

  2. bradygaster commented on Aug 25, 2026

    @bradygaster
    CollaboratorAuthor

    Caution

    Flight readiness assessment — Phase 2 (#1734)

    Decision: NO-GO · Assessed: 2026-08-25 · P1 evidence gate only; this does not authorize enforcement work.

    The phased rollout in #1729 and the acceptance gate in #1734 require live advisory evidence showing that false REQUEST_CHANGES verdicts will not stall the pipeline. Phase 1 merged only about 15 hours ago in #1871. No production /squad review runs or submitted advisory verdicts are evidenced in this repository, and no linked consumer-run artifacts were found. The false-positive and false-negative rates are therefore unknown (0/0), not passing.

    Evidence ledger

    Area Finding Gate status
    Advisory source and authority contract #1871 shipped the read-only reviewer contract; structural/mutation tests support the design. PASSING EVIDENCE (static only)
    Install/docs surface #1874 documented consumer installation. #1875 is a follow-up cleanup, not soak evidence. PASSING EVIDENCE (shipping only)
    ready_for_review No qualifying post-ship agent-authored PR was found in bradygaster/squad; no live firing to inspect. UNKNOWN — absence of evidence
    synchronize No live firing, re-delivery, or head-SHA transition was found. UNKNOWN — absence of evidence
    Manual /squad review No post-ship manual review invocation or resulting run/review was found here. UNKNOWN — absence of evidence
    Provenance attribution Durable marker, squad/implement-*, Copilot-bot, and manual-unattributed paths are covered structurally but have not been observed live. UNKNOWN — absence of evidence
    SHA deduplication No repeated-trigger or same-head noop artifact exists. UNKNOWN — absence of evidence
    COMMENT vs REQUEST_CHANGES Allowed outcomes are constrained statically, but no real advisory verdict exists to assess semantic correctness. UNKNOWN — absence of evidence
    False positives / false negatives No emitted verdicts and no human adjudications against them. UNKNOWN — 0/0, not a pass
    Authorization boundaries Read-only permissions and advisory-only safe outputs pass static checks; no authorized/unauthorized manual-trigger pair or fork refusal has been exercised live. PASS static / UNKNOWN live
    Agent-authored PR outcomes No qualifying agent-authored PR opened after #1871, so none received an expected advisory verdict. UNKNOWN — no subject population

    Distribution-model clarification

    bradygaster/squad distributes workflows/squad-review.md; consumer repositories install and compile it with gh aw add. The absence of an installed Squad Review workflow under this repository's .github/workflows/ is intentional and is not itself a defect. It does mean this repository cannot provide in-repo soak unless it self-installs the workflow. Any soak in consumer repositories is currently unknown because no concrete consumer run/review URLs are linked from #1729, #1734, #1871, #1874, or #1875.

    Blocking decision

    NO-GO for Phase 2 enforcement. Green unit tests and compiled-contract checks establish that the workflow is shaped safely; they do not establish trigger reliability, verdict calibration, deduplication, live authorization behavior, or agent-PR coverage.

    Minimum additional soak before reassessment

    • At least 2 weeks of continuous advisory operation in one or more identified consumer installations.
    • At least 10 submitted advisory reviews across 5 qualifying agent-authored PRs, with exact run, PR, and review URLs retained.
    • Live coverage of both automatic triggers: at least one ready_for_review and two synchronize runs on distinct head SHAs.
    • At least two authorized manual /squad review invocations, plus one unauthorized invocation showing rejection/noop.
    • At least one same-head repeated trigger proving dedup/no duplicate review.
    • At least one fork or unattributed automatic trigger proving fail-closed/noop behavior.
    • Human adjudication of every REQUEST_CHANGES, with false-positive rate below the agreed enforcement threshold (proposed: <10%) and at least one known blocker caught to provide a false-negative signal.
    • Evidence that qualifying agent-authored PRs receive the expected advisory verdict and Squad-Review-Head: attribution.

    Reassess only when those artifacts are linked here. Until then, absence of incidents is absence of exposure—not readiness evidence.

  3. bradygaster commented on Aug 30, 2026

    @bradygaster
    CollaboratorAuthor

    Evidence gate (Ralph triage): This phase is explicitly blocked on an advisory soak of the Phase 1 MVP (#1733) and on the Tier 1 false-green fixes landing — a required check backed by blind instrumentation is equivalent to no gate. #1828 (reviewer lockout enforcement surface) is a direct architecture/decision dependency: the enforcement mechanism cannot be designed until the lockout protocol question is resolved. Priority adjusted from p1 to p2 to reflect the blocked state. \go:needs-research\ left in place — no existing blocked/evidence label to replace it with without creating a new label.

  4. github-actions commented on Sep 10, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  5. github-actions commented on Sep 10, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  6. github-actions commented on Sep 11, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  7. github-actions commented on Sep 11, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  8. github-actions commented on Sep 11, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  9. github-actions commented on Sep 11, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: Flight
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  10. github-actions commented on Sep 13, 2026

    @github-actions
    Contributor

    🔄 Ralph — Auto-Triage

    Assigned to: architect
    Reason: No module, routing, or role keyword match — routed to Lead/Architect
    Source: lead-fallback

    Ralph auto-triaged this issue using routing rules.
    To reassign, swap the squad:* label.

  11. added
    squad:agentic-workflows-devAssigned to agentic-workflows-dev (Agentic Workflows Engineer)
    and removed
    squad:architectAssigned to architect (Lead Architect)
    on Sep 24, 2026
  12. github-actions commented on Sep 24, 2026

    @github-actions
    Contributor

    📋 Assigned to agentic-workflows-dev (Agentic Workflows Engineer)

    Issue: #1734 — Phase 2: Enforce /squad review as a required check and gate epic relay on verdict

    agentic-workflows-dev will pick this up in the next Copilot session.

    For Copilot coding agent: If enabled, this issue will be worked automatically.
    Otherwise, start a Copilot session and say:
    agentic-workflows-dev, work on issue #1734

  13. bradygaster commented on Sep 28, 2026

    @bradygaster
    CollaboratorAuthor

    Implementation is tracked by #2096, whose closing keyword has been changed to Implements #1734. Keep this issue open after merge until a private consumer fixture proves the exact-head Squad Review check, logical author/reviewer separation, relay gating, SHA deduplication, and the documented rollout/soak gate.

  14. added 2 commits that reference this issue on Sep 29, 2026
    ad8a0f6
    b282987
  15. bradygaster commented on Sep 29, 2026

    @bradygaster
    CollaboratorAuthor

    Implementation milestone: #2096 merged into dev as da7861d after exact-head independent review, CLEAN state, and required test + gh-aw strict compile success. Keep this issue open until the private consumer full-lifecycle fixture records the exact-head Squad Review check, logical reviewer separation, relay/deduplication behavior, and the documented rollout/soak.

  16. bradygaster commented on Sep 29, 2026

    @bradygaster
    CollaboratorAuthor

    Reopened because merge automation closed this issue before the documented consumer evidence and rollout/soak gates were satisfied. The implementation is merged; hosted acceptance and policy rollout remain.

  17. bradygaster commented on Sep 29, 2026

    @bradygaster
    CollaboratorAuthor

    Post-#2105 reconciliation

    The stable Squad Review / review contract and fail-closed dedicated-App design are merged through 29e69f53. Closure still requires the real dedicated integration ID, integration-bound required check, hosted relay/dedupe/override proof, and soak. Those cannot begin until #2103 provisions squad-review-authority and the reviewer App. Keep open; the earlier auto-closure was premature.

  18. added a commit that references this issue on Oct 20, 2026
    c24da2a
  19. added
    wave:1-nextWave 1 — immediate next sprint
    wave1:P8Wave 1 PR group P8
    and removed
    wave:3-laterWave 3 — backlog, future
    on Nov 5, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    gh-awGitHub Agentic Workflows integrationpriority:p2Next sprintsquadSquad triage inbox — Lead will assign to a membersquad:agentic-workflows-devAssigned to agentic-workflows-dev (Agentic Workflows Engineer)triagedIssue has been triagedtype:featureNew capabilitywave1:P8Wave 1 PR group P8wave:1-nextWave 1 — immediate next sprint

    Type

    No type

    Projects

    No projects

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions