review-preflight

===SKILL:REVIEW_PREFLIGHT=== META: TYPE::SKILL VERSION::"1.2.0" STATUS::ACTIVE PURPOSE::"Mechanical prep delegation before review. Collects structured context brief via pal_chat so CRS spends tokens on judgement, not grunt work."

§1::CORE AUTHORITY::ADVISORY[preflight_brief_structure] PHASE::PRE_REVIEW[runs_BEFORE_review-discipline⊕review-prioritization] COMPLEMENTS::[review-discipline,review-prioritization]

§2::PROTOCOL // CRS delegates preflight collection via mcp__pal__chat // CRS runs as subagent (no Task tool) — PAL MCP tools are the delegation path DETECT_CONTEXT::[ PR::"gh_pr_view→has_diff⊕has_CI⊕has_tier", STANDALONE::"file_list_provided→no_PR_metadata", AUDIT::"explicit_audit_scope→security_focus" ] COLLECT::[ CONTEXT_TYPE::PR∨STANDALONE∨AUDIT, FILE_SUMMARY::"files_changed⊕line_delta⊕languages_touched", CI_STATUS::"pipeline_result[PR_only]", TIER::"T0_through_T4_classification[PR_only]", SECURITY_PATHS::"auth∨crypto∨secrets∨permissions∨.env∨tokens", TEST_MAP::"changed_files→existing_test_files[presence_only]", BOT_FINDINGS::"extracted_bot_review_comments[PR_only∧ADVISORY]" ] // BOT_FINDINGS extraction protocol BOT_AUTHORS::[ PRIORITY_1::cubic-dev-ai[bot]<structured_confidence⊕P0-P2_tiers⊕agent_prompts>, PRIORITY_2::github-copilot[bot]<inline_suggestions∧auto_review> ] // LOGIN_NORMALIZATION: GitHub APIs return different login formats for the same // bot accounts. gh pr view --json comments returns logins WITHOUT the [bot] // suffix (e.g., "cubic-dev-ai"), while gh api // repos/{owner}/{repo}/pulls/{pr}/comments preserves the [bot] suffix (e.g., // "cubic-dev-ai[bot]"). Additionally, some bots use legacy login variants // that differ from their canonical name: // - github-copilot[bot] may appear as "Copilot" or "copilot" (no [bot] suffix) // - cubic-dev-ai[bot] may appear as "cubic-bot" // - github-actions is the CI automation account (not a [bot]-suffixed account); // its comments are excluded from gate validation because CI status comments // contain approval-like text that would falsely clear review gates. // The validate_review.py BOT_LOGIN_SET handles this by stripping [bot] from // ADVISORY_BOTS and adding known legacy variant logins. When scanning for bot // comments, match against the normalized set, not the canonical names. BOT_EXTRACT::[ SCAN_ISSUE_COMMENTS::"gh_pr_view--json_comments→filter_by_normalized_BOT_LOGINS[strips_bot_suffix⊕includes_github-actions]", SCAN_REVIEW_COMMENTS::"gh_api_repos/{repo}/pulls/{pr}/comments→filter_by_NORMALIZED_BOT_LOGINS[match_user.login_against:cubic-dev-ai|cubic-bot|github-copilot|Copilot|copilot|github-actions⊕also_match_[bot]_suffix]", CUBIC::"extract_P0_P1_findings⊕confidence_metadata⊕agent_prompt_sections", COPILOT::"extract_inline_suggestions[ADVISORY_context_only]", CLASSIFY::ADVISORY[bot_findings_NEVER_block_merge] ] OUTPUT::STRUCTURED_BRIEF[consumed_by_CRS⊕CE_before_review_begins]

§3::GOVERNANCE DELEGATION::[ TOOL::mcp__pal__chat[or_mcp__pal__clink_for_CLI_providers], PROVIDER_AGNOSTIC::no_model_specification[runtime_selects], FILE_PATHS::pass_absolute_file_paths_parameter[not_inline_code] ] BUDGET::[ BRIEF_ONLY::collect_facts_not_judgements, NO_REVIEW::delegate_NEVER_evaluates_code_quality, CONCISE::brief_fits_in_500_tokens_max ]

§5::ANCHOR_KERNEL TARGET::structured_preflight_brief_with_bot_findings_before_review NEVER::[evaluate_code_quality_in_preflight,specify_model_or_provider,skip_context_detection,produce_findings_or_verdicts,treat_bot_findings_as_blocking] MUST::[detect_review_context_type,collect_file_summary_and_languages,map_security_sensitive_paths,map_changed_files_to_tests,scan_both_issue_comments_AND_review_comments_for_bot_findings,normalize_bot_logins_before_matching,classify_bot_findings_as_ADVISORY,delegate_via_pal_tools] GATE::"Is a structured preflight brief (including bot findings marked ADVISORY) collected via PAL delegation before CRS begins review judgement?"