8.9 KiB
| template_type | runtime_id | agent_id | version | status |
|---|---|---|---|---|
| local-ccra-review-report | ccra-local-review | ccra-local-reviewer | 0.1.2 | draft |
Local CCRA Review Report
1. Return Metadata
- project_id:
- review_round:
- local_pass:
- turn_type: review_turn
- runtime_id: ccra-local-review
- runtime_version:
- agent_id: ccra-local-reviewer
- agent_version:
- invocation_mode:
- carrier_type:
- child_session_id:
- persistent_available:
- execution_status: returned
- simulation_only: false
- report_path:
- returned_output_path:
- bundle_audit_profile:
- gate_execution_mode: dry_run | run | owner_waived | not_applicable
- routing_diff_policy: required | skipped_no_after_evidence | manual_equivalent_required | not_applicable
- lifecycle_scan_scope: current_review_bundle | changed_files | knowledge_assets | model_library_state | whole_repo
- generated_at:
2. Review Scope
- primary_question:
- scope_statement:
- non_goals:
- local_tuning_stage:
- web_final_review_target:
- knowledge_base_path:
- previous_turn_report_path:
3. Inputs Actually Received
List the actual files, directories, helper outputs, and prior decisions received. Mark missing expected inputs explicitly.
4. Files Actually Read
Record exact paths or source-relative paths.
- path:
purpose:
notes:
5. Automation Outputs Used
- helper_id:
output_path:
status: available | unavailable | planned | executed | skipped | waived | failed | dry_run | stale | generic_profile | profile_mismatch | broad_scope_noise | missing_helper
active_profile:
recognized_manifest_path:
gate_status: planned | executed | skipped | waived | not_applicable
blocks_current_review: yes/no
how_used:
6. Executive Summary
State the review result in practical terms:
ready_for_local_tuning
needs_more_local_tuning
ready_for_web_final_review_after_owner_approval
blocked_by_missing_evidence
blocked_by_scope_or_lifecycle_risk
Then state the compact final action recommendation:
fix_now
document_only
defer_to_calibration
do_not_open_patch_round
ready_for_web_review_after_local_repairs
6.1 Short Returned-Output Summary
Copy this compact shape into returned-output.md so the Owner/main session can read the return without parsing the full report.
review_round:
local_pass:
turn_type:
child_session_id:
main_conclusion:
final_action_recommendation:
blocks_web_review: yes/no
blocks_owner_acceptance: yes/no
required_next_action:
output_paths:
review_report:
planning_output:
comparison_output:
evidence_limitations:
- limitation:
7. Issue Mechanism Diagnosis
For each material issue, diagnose mechanism before recommending action.
Issue ID:
Case / Material:
What failed:
Why it failed:
Why it is not a different failure category:
Failure category:
product_routing_problem | rule_boundary_problem | model_content_problem | runtime_problem | lifecycle_governance_problem
Evidence:
For selector and routing issues, explicitly distinguish:
rule-based selector failure
LLM understanding failure
instruction / payload boundary failure
model-level exclusion failure
depth-limiting override failure
no-call false negative
model over-selection
model under-selection
8. Patch Scope Judgment
Separate cases and recommendations:
must_fix_cases:
- case_id:
reason:
required_evidence:
same_class_boundary_cases:
- case_id:
reason:
patch_with_current_round: yes/no
calibration_only_cases:
- case_id:
reason:
explicitly_not_patched:
- case_id:
reason:
should_not_open_another_small_patch_round:
- issue_or_case:
reason:
9. Verification Evidence Requirements
When a patch is proposed or already exists, state whether the package contains enough verification evidence.
Required for selector or routing patches:
targeted_regression_for_confirmed_target_cases:
full_frozen_pool_rerun_or_before_after_comparison:
existing_selector_regression:
calibration_smoke:
unit_tests:
command_log_or_validation_report:
gate_execution_mode: dry_run | run | owner_waived | not_applicable
gate_evidence_status: planned | executed | skipped | waived | not_applicable
dry_run_vs_executed_gate_status:
rule_diff_only_is_sufficient: no
missing_evidence:
Dry-run output may be cited as planning/preflight evidence only. It must not be used as a passing validation result in the final action recommendation.
10. Before / After Behavior Review
Required whenever post-patch or before/after evidence exists.
target_behavior_changes:
- case_id:
before:
after:
assessment:
non_target_behavior_changes:
- case_id:
before:
after:
assessment:
accepted_collateral_behavior_changes:
- case_id:
rationale:
regression_or_documentation:
unchanged_expected_behavior:
- case_id:
evidence:
assessment:
unexpected_regressions:
- case_id:
evidence:
action:
cases_needing_owner_or_web_calibration:
- case_id:
question:
If before/after evidence exists but routing-behavior-diff-audit was skipped, state that as helper-output confidence risk.
If the helper cannot run, include a manual equivalent table covering all categories above.
11. Non-Target Behavior Changes
Record all observed non-target or collateral changes, including cases that are acceptable but must be documented.
12. Calibration-Only Candidates
List cases or categories that should be deferred to Owner/Web CCRA calibration rather than patched immediately.
- case_or_category:
why_calibration_only:
risk_if_patched_now:
owner_or_web_question:
13. Helper Output Confidence
For each helper:
helper_id:
output_used_as_evidence:
confidence: high | medium | low | not_used
status: planned | executed | dry_run | skipped | waived | failed | stale | generic_profile | profile_mismatch | false_positive_candidate | broad_scope_noise | missing_helper
active_profile:
recognized_manifest_path:
missing_required_files:
extra_files_or_profile_warnings:
confidence_notes:
blocks_current_review: yes/no
requires_runtime_tuning: yes/no
Must distinguish:
helper output used as evidence
helper output that appears stale, generic, false-positive, or profile-mismatched
dry-run versus executed validation
missing helper that should have run
generic bundle-audit profile mismatch versus real missing review material
broad whole-repo lifecycle noise versus current-scope blocker
whether helper findings block the current review or only require runtime tuning
14. Findings
Use one entry per finding.
Finding ID:
Classification: blocker | major | minor | calibration | defer
Title:
Evidence:
Why It Matters:
Recommended Action:
Final Action Recommendation: fix_now | document_only | defer_to_calibration | do_not_open_patch_round | ready_for_web_review_after_local_repairs
Owner Decision Required: yes/no
Blocks Web Final Review: yes/no
Blocks Lifecycle Change: yes/no
15. Engineering Pass vs Product Judgment
Separate:
What engineering evidence shows:
What product/model-governance question remains:
What evidence would close the gap:
16. Lifecycle And Scope Guard
Record any status, stability, acceptance, round closure, or lifecycle-promotion claims found during review.
scan_scope: current_review_bundle | changed_files | knowledge_assets | model_library_state | whole_repo
Claim:
Evidence:
Required authority:
Triage:
current_scope_blocker | owner_decision_needed | historical_or_archive_noise | policy_reference_not_claim | false_positive_or_low_confidence
Assessment:
Do not turn historical review chatter, archived helper outputs, or policy examples into current lifecycle blockers unless the report explains how the claim affects the current review scope.
17. Local Tuning Recommendations
Prioritize actions for the project main session.
1.
2.
3.
18. Web Final Review Readiness
Choose one:
not_ready
ready_after_local_repairs
ready_after_owner_decision
ready_for_web_final_review
Explain why.
19. Final Action Recommendation
Choose the highest-level recommendation for the reviewed material:
fix_now
document_only
defer_to_calibration
do_not_open_patch_round
ready_for_web_review_after_local_repairs
State what should happen next and what should not happen next.
This recommendation must not rely on dry-run gate output as pass evidence.
20. Owner Decision Points
- decision_id:
decision_needed:
options:
recommended_default:
21. Limits And Uncertainty
State what was not checked and why.
22. Resume Instructions
If the same child session should continue, state the next turn type.
For a planning turn, provide:
turn_type: planning_turn
same_child_session_id:
first_turn_report_path:
owner_decision_path:
next_review_target:
planning_output_path: next-review-requirements.md
If a new child session is needed, list the files that must be included and state that same-session continuity was not preserved.