[Agent Lock] Block completion unless intent, diff, checks, and output agree

[groupthinking]

Objective

Add a deterministic Agent Completion Truth Gate that evaluates repository evidence instead of trusting agent completion prose.

Agent login

groupthinking

Agent run ID

codex-myx-82-20260717

Incident fixture

PR #813 / issue #802 exposed the required failure modes:

• the source issue had no description or acceptance criteria;
• Jules emitted both "Ready for review" and "unexpected error / wasn't able to complete";
• the PR claimed two source files but changed five files across three commits;
• unrelated dependency and test changes entered the branch;
• generic CI was green while an unresolved review identified a behavior-breaking regex regression;
• no focused regression test demonstrated the intended behavior.

Required behavior

The gate must return a machine-readable verdict with explicit reasons and fail closed.

Pre-dispatch

• Reject an issue with a blank description.
• Reject an issue without explicit acceptance criteria.
• Require a declared file scope or an explicit unrestricted-scope marker.

Post-execution

• Compare declared file scope with actual changed paths.
• Treat any undeclared path as scope drift unless explicitly allowlisted.
• Treat contradictory completion and error events as failed.
• Treat unresolved blocking review findings as failed even when generic CI is green.
• Require focused test evidence for behavior-changing source edits.
• Only allow a completion verdict after the PR is merged and required post-merge checks succeed.

Deliverables

• A small deterministic validation module with no external service dependency.
• A CLI suitable for GitHub Actions.
• Unit tests using PR 🧹 [code health] Obfuscate literal TODO strings in application logic and templates #813-shaped fixtures.
• A GitHub Actions workflow that runs on pull requests and uploads/writes the verdict as a check artifact.
• Documentation of the input schema, verdict schema, and fail-closed rules.

Declared file scope

• scripts/ci/agent_completion_gate.py
• tests/unit/test_agent_completion_gate.py
• tests/fixtures/agent_completion/pr_813.json
• .github/workflows/pr-checks.yml
• .github/ISSUE_TEMPLATE/agent-task.yml
• .github/pull_request_template.md
• docs/agent-completion-truth-gate.md

Focused test paths

• tests/unit/test_agent_completion_gate.py

Acceptance tests

• Blank issue description returns blocked.
• Missing acceptance criteria returns blocked.
• Declared two files but actual five returns blocked: scope_drift.
• Completion plus later error returns blocked: contradictory_terminal_events.
• Green CI plus unresolved blocking review returns blocked: unresolved_review.
• Source behavior change without focused test evidence returns blocked: missing_test_evidence.
• Unmerged PR cannot return completed.
• A fully scoped, reviewed, tested, merged fixture returns completed.

Pre-dispatch confirmation

• Objective, acceptance criteria, scope, and focused-test expectations are complete before delegation.

Non-goals

• No autonomous merging.
• No LLM judgment inside the enforcement path.
• No Slack or notification layer.
• No modification of application runtime behavior.

Related: #813, #802.

[linear-code]

MYX-82

[github-actions]

👋 Thanks for opening this issue! It has been automatically labeled for triage.

[coderabbitai]

🔗 Related PRs

#277 - Vercel remediation complete: waitUntil adoption + active middleware + @vercel/functions package (resolves three user-defined outcomes; 16-agent verification-gate consolidated PASS on full matrix) [closed]
#404 - fix(web): sync package-lock with @vercel/functions (unblocks #277 auto-merge) [merged]
#411 - ci: add agentic verification loop workflow and refactor state log [merged]
#593 - ci(phase-goal-tracker): goal tracker workflow with reviewer findings resolved (supersedes #531) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!