القائمة
Run 4 lightweight quality checks on recent work and fix issues found
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | evaluate |
| description | Run 4 lightweight quality checks on recent work and fix issues found |
Run 4 quality checks on recently modified files. Identify issues. Fix them with user approval.
Before running checks, determine review level based on change scope:
Lines changed < 50 AND files <= 2 → Level 1-2 (self-check + peer)
Lines changed 50-200 AND files 2-5 → Level 1-3 (+ devil's advocate)
Lines changed 200-500 AND files 5-10 → Level 1-5 (+ adversarial debate)
Lines changed > 500 OR files > 10 → Level 1-7 (full bridge briefing)
Override minimums for sensitive files:
After running the 4 checks below, suggest the appropriate next review level based on this scaling.
flowchart TD
Start([evaluate]) --> GetFiles["Step 0: git diff --name-only HEAD\n+ git diff --name-only --cached"]
GetFiles --> HasFiles{Files found?}
HasFiles -- No --> AskUser[Ask user to specify files]
AskUser --> HasFiles
HasFiles -- Yes --> Scale0["Assess review level\n(lines changed + file count + sensitivity)"]
Scale0 --> C1["Check 1: Code Quality\nlint-check.sh"]
C1 --> C2["Check 2: Goal Alignment\nmodel + memory"]
C2 --> C3["Check 3: Tests\ntest-runner.sh"]
C3 --> C4["Check 4: Security\nOWASP Top 10"]
C4 --> Table[Output results table]
Table --> Fail{Any FAIL or WARN?}
Fail -- No --> Scale[Suggest next review level]
Scale --> Done([Done])
Fail -- Yes --> InitCycle["cycle = 1"]
InitCycle --> Fix["List fixes + diff preview"]
Fix --> Approve{"Cycle cycle/3:\nApply fixes?"}
Approve -- No --> Report[Report remaining issues]
Report --> Done
Approve -- Yes --> Apply[Apply fixes]
Apply --> ReRun[Re-run ONLY failed checks]
ReRun --> Still{Still FAIL/WARN?}
Still -- No --> Scale
Still -- "Yes + cycle < 3" --> IncCycle["cycle++"] --> Fix
Still -- "Yes + cycle = 3" --> Escalate[Escalate to human review]
Escalate --> Done
Check 1 -- Code Quality: Run ~/.claude/scripts/lint-check.sh <modified-files>. Parse JSON: PASS if status == "pass", FAIL with each issues item.
Check 2 -- Goal Alignment: Query mcp__memory__search_nodes for entities related to modified files/modules before evaluating. Verify: (1) code matches request, (2) no unrequested additions, (3) no missing requirements, (4) aligns with knowledge graph decisions. Report ALIGNED or MISALIGNED with specific divergences.
Check 3 -- Integration/Tests: Run ~/.claude/scripts/test-runner.sh. Parse JSON: PASS if status == "pass", FAIL with failures array + counts.
Check 4 -- Security: Scan modified files only for OWASP Top 10:
Report PASS or WARN with file:line references.
After all 4 checks, output a summary table:
## Evaluate Results
| Check | Status | Notes |
|-----------------|--------|--------------------------------|
| Code Quality | PASS | eslint: 0 issues |
| Goal Alignment | PASS | All requirements addressed |
| Tests | FAIL | 2 failures: test_auth, test_db |
| Security | WARN | Possible XSS at src/render.js:42 |
Escalation message after 3 cycles: "These issues require human review after 3 fix cycles: [list]"
Scan for stale memory topics, near-dead instincts, and low-salience K-LEAN entries. Proposes cleanup actions interactively.
Quality gate for instinct persistence. Validates new instincts through a 4-stage pipeline before allowing them to be written to disk. Invoke before writing any new instinct file.
Use when the user needs thorough research on a complex topic, says 'deep dive', 'research this thoroughly', or 'sensor sweep'. Parallel agent orchestrator for comprehensive multi-angle investigation.
Autonomous experiment loop: iteratively improves a single file against a measurable metric. Uses git-as-state-management in a disposable worktree. Invoke when user wants to optimize, tune, or systematically improve a file through repeated iterations.
Use when you need multi-faction review combining security, strategy, cost, and quality perspectives. Trigger: fleet review, multi-faction assessment, comprehensive quality gate, or when escalation requires L5+.
Two-loop research orchestrator: chains deep-research (outer, hypothesis generation) with experiment-loop (inner, empirical testing). Use when research questions have a testable dimension with a measurable metric.