Autonomous Review And QA Evaluation

Contracts

Read contracts/index.yaml first; do not load every contract. For entry, load contracts/input.yaml review-entry fields selected by review-entry-fields in contracts/index.yaml.

• Triggered change_impact and architecture_pack_input: resolve before planning or dispatch; load the active round from contracts/phase-gates.yaml at each transition, and compact contracts/handoffs.yaml pointer before Reviewer/QAEvaluator dispatch. Resolve reviewer_context at pass start, return_validation after results, and contracts/output.yaml before exit; keep change-impact guidance out of Reviewer bundles.

Migration note: assistant-review contracts are v8.0. Persisted 6.0/7.0/7.1/7.2 batch packets are invalidated and rebuilt from a fresh 8.0 snapshot before CLEAN. Preserve recoverable selected design, rationale, and viable alternatives/dispositions. Workflow-composed no_build_scope is allowed only when no fix is required or occurred and no due selected or applicable due/executable binding check exists; every due check needs current passed evidence. After a fix, current passed evidence is mandatory. Standalone review-fix still requires successful build and tests. Apply subagent_trigger_scope; opt-out, unavailability, or policy blocks use direct fallback. Reviewer returns and final summaries require non-empty reviewed_scope so workflow consumers can use the producer packet.

Selectors resolve canonical fields; entry loads no review guidance. For a missing/invalid selector, use load_full_authoritative_file, validate the full canonical file, and record recovery.

Goal

Find and fix authorized evidence-backed defects, regressions, and test gaps; report calibrated findings.

Success Criteria

• Resolve scope, mode, and material before the loop; rank findings by severity, evidence, and confidence.
• Each frozen-snapshot batch plans and attempts at least two independent passes. All expected passes reach terminal accounting before aggregation, fix, or exit. CLEAN/ISSUES_FIXED require complete coverage. Audit exits after one started batch with non-empty reviewed_scope; failed, timed-out, blocked, or otherwise incomplete coverage returns HAS_REMAINING_ITEMS.
• Apply SOLID, KISS, DRY, YAGNI, and readability from references/review-principles.md; validate every applicable Pack, including invalid ones, with its Review Checklist.
• Review-fix resolves or explicitly defers must-fix/should-fix findings, validates, then re-reviews. Required QA returns evidence-backed score progression and final acceptance.
• Workflow-composed fixes enforce F_REVIEW_VERIFICATION_SELECTION.
• Select qa_evaluation_mode from contracts/input.yaml. QA required positive triggers: explicit QA/acceptance evaluation request, accepted Done Contract, harness-capable acceptance scope, domain-scored scope, or scoped UI/visual/product/UX/docs/DX acceptance. QA non-triggers: template labels/placeholders, generic acceptance criteria labels, optional/not_required reasons, delegation/source-changing work alone, and ordinary medium+ code-review-only/source-changing work.
• Load references/domain-rubrics.md only when acceptance criteria, Done Contract, domain_context, or explicit rubric_refs scope domain quality; return selected_domain_rubrics/domain_quality_scores when scoped.

Constraints

• Infer edit authority: report-only is audit; clear source authorization is review-fix; carried approval permits bounded in-scope fixes. If audit versus editing is materially unresolved, ask exactly: "Should I only report findings, or also implement and verify fixes?" Do not review/mutate until answered.
• Keep QA evaluation separate from code review: QA owns acceptance, Done Contract, verification, scoped quality, progression, and final result. Code Reviewer still owns code defects, security, architecture, and test-coverage review.

Entry

Review explicit material, then uncommitted changes, journal/packet, or current files; ask only if material is unavailable.

A standalone review this runs Spec Review on user scope. Review-fix repairs authorized mismatches and repeats to PASS before Reviewer dispatch; audit retains the Spec Review mismatch as an aggregate finding and continues a frozen read-only multi-pass batch. The task journal is optional. A workflow-composed review consumes carried Spec Review PASS and current passed verification for every due selected CheckSpec and resolved binding check. After any source fix, each subsequent Reviewer dispatch requires fresh current passed selected/binding evidence; applicable builds/tests must pass and not_applicable admission is forbidden.

Review Modes

Select applicable spec, regression, test, maintainability, bugfix-evidence, semantic/behavioral contract, agentic-loop safety, and security modes. Three entry flags enable contract/loop modes; route security to Code Reviewer with the assistant-security checklist/perspective. risk_selected_specialist uses Code Reviewer with the selected specialist's checklist/perspective and the canonical Reviewer schema.

Ownership and environment inform material, never edit authority. Return out-of-scope findings to workflow. Report severity (must-fix, should-fix, nit), file/line evidence, impact, smallest useful fix, and calibrated confidence; speculation is a non-blocking Observation.

QA evaluation runs after code-review/build evidence. Load references/qa-evaluation-loop.md when qa_evaluation_mode=required.

Company-Safe Review Rules

Use local diffs/repo-native checks; external scans/reviews or installs need approval. Redact secrets/proprietary data; offer local/manual alternatives if blocked.

Mandatory Review Checklists

The fresh Reviewer context bundle points to references/review-checklists.md; applicable sections yield findings or an explicit "no concrete risk found" check. Agentic loop flag -> Agentic Loop Safety Checklist -> agentic_loop_safety_checks; map Behavioral Contract Review Checklist/behavioral_contract_checks; Semantic Contract Review Checklist/semantic_contract_checks; Architecture Decision Pack Review Checklist/architecture_decision_pack_checks.

Refactor-Related Findings

Allowed risk categories: correctness/security, unsafe surface, branching/responsibility growth, hidden dependency/ownership, brittle tests, extension seams, and readability/maintainability. Every refactor-related finding MUST state its risk category, affected surface, evidence, and smallest durable fix. Use concrete risk framing instead of generic convention, style, cleanliness, or improvement language. Request broad cleanup only when a smaller durable fix cannot remove the risk.

Architecture Decision Pack Review

Review each applicable Pack or equivalent ADR, even if invalid; validate and report it. Match its mode to canonical architecture_design_mode; require independent challenge evidence for review_intensive. Apply the full Architecture Decision Pack Review Checklist in references/review-checklists.md and report gaps.

Principle and Readability Lens

For medium+ reviews, run the Design Coherence Pass and return evidence-bound principle_checks.design_coherence; findings name surface, risk, and smallest fix.

Review Loop Routing

After entry, load references/review-loop.md before the first REVIEW step. It owns batches, transitions, barriers, and pivots. Fresh pass bundles carry principles/checklists/rubrics. worker_return_schema_selector resolves only recursive required/triggered shapes, types, enums, and cardinality; exclude sibling/batch state. Impact projection preserves canonical closure and cannot make audit/review-fix clean.

For a carried approved_feature_preparation_qa_acceptance_obligation, verify its scope, prerequisite, source binding, and exact result/successful-pair conditions with evidence from contracts/handoffs.yaml and contracts/phase-gates.yaml; return rejected/HAS_REMAINING_ITEMS or blocked/BLOCKED when unmet.

Exit: Present Final Result

For no findings, say: "No material findings within the reviewed scope and available evidence"; CLEAN is an enum, not proof.

Rules

• Keep round results internal in fresh sibling-blind bundles; incomplete coverage returns HAS_REMAINING_ITEMS. Verify closure of prior fixes.

Output

Return reviewed scope, rounds/result, evidence-backed findings/fixes, verification, applicable bugfix/agentic/behavioral/semantic checks, required QA, and residual risk per contracts/output.yaml.

Stop Rules

• Audit mode stops after one review batch. Report terminal findings without edits.
• Review-fix runs round 1, authorized fixes/validation, then a fresh round 2 (max 10 rounds); mutation requires a new snapshot and complete coverage.
• Before round 3+, require additional_round_reason backed by changed files, unresolved finding, validation failure, regression/drift, or changed hypothesis; score alone is insufficient. round 10 is terminal; never start 11.
• Block and stop for unavailable/empty required review material.

Drift detection (medium+ scope)

Compare rounds with references/score-tracking.md. Drift, regression, stagnation, or pivot evidence returns pivot_restart_signal; the orchestrator records pivot_restart_decision before another pass.

Review Finding Rule Distillation

For each blocker/must-fix, load references/review-finding-permanent-rule.md; classify one_off_fix, permanent_rule_candidate, or no_action. Promote only recurring process/eval gaps, missing contracts/checklists, or high-impact repeat failures.