Codex Iterative Fix

Codex Iterative Fix Skill

Kanitsal Cerceve (Evidential Frame Activation)

Kaynak dogrulama modu etkin.

Purpose

Use Codex CLI's autonomous iteration capability to fix failing tests and debug issues until they pass. This skill leverages Codex's strength at long-horizon coding tasks.

When to Use This Skill

• Multiple tests failing that need iterative fixing
• Debugging issues requiring trial-and-error
• Refactoring with test validation
• CI/CD failures needing automated fixes
• Build errors requiring multiple attempts

When NOT to Use This Skill

• Research tasks (use multi-model-discovery)
• Understanding codebase (use gemini-codebase-onboard)
• Critical production code (use sandbox mode first)
• Architecture decisions (use llm-council)

Workflow

Phase 1: Initial Assessment

1. Identify failing tests/errors
2. Determine scope of fixes needed
3. Choose appropriate mode:
   • full-auto: Standard autonomous mode
   • sandbox: For risky changes
   • yolo: When speed is critical

Phase 2: Codex Execution

# Standard iterative fix
./scripts/multi-model/codex-yolo.sh "Fix all failing tests" task-id "." 15 full-auto

# With sandbox protection
./scripts/multi-model/codex-yolo.sh "Fix tests" task-id "." 10 sandbox

# Via delegate.sh
./scripts/multi-model/delegate.sh codex "Fix all failing tests and verify they pass" --full-auto

Phase 3: Verification

1. Codex runs tests after each fix
2. Iterates until all tests pass
3. Claude reviews final changes
4. Summary of what was fixed

Success Criteria

• All targeted tests passing
• No new regressions introduced
• Changes reviewed and validated
• Clear documentation of fixes

Example Usage

Example 1: Test Suite Failures

User: "CI is failing with 12 test errors"

Codex Process:
1. Run tests, capture failures
2. Analyze first failure
3. Implement fix
4. Re-run tests
5. Repeat until all pass

Output:
- Fixed: 12 tests
- Changes: 8 files modified
- Root causes: Missing null checks, outdated mocks

Example 2: Type Errors

User: "TypeScript build has 47 type errors"

Codex Process:
1. Run tsc, capture errors
2. Fix type errors systematically
3. Re-run after each batch
4. Verify build succeeds

Output:
- Fixed: 47 type errors
- Patterns: Missing types, incorrect generics
- Added: 3 new type definitions

Modes Explained

Integration with Meta-Loop

META-LOOP IMPLEMENT PHASE:
    |
    +---> codex-iterative-fix
    |         |
    |         +---> Codex: Run tests
    |         +---> Codex: Fix failures
    |         +---> Codex: Iterate until pass
    |
    +---> Continue to TEST phase (verification)

Memory Integration

Results stored at:

• Key: multi-model/codex/iterative-fix/{project}/{task_id}
• Tags: WHO=codex-iterative-fix, WHY=autonomous-fixing

Invocation Pattern

# Via router (automatic detection)
./scripts/multi-model/multi-model-router.sh "Fix all failing tests"

# Direct Codex call
bash -lc "codex --full-auto exec 'Fix all failing tests and verify they pass'"

Guardrails

NEVER:

• Run on production without review
• Skip final verification
• Ignore new test failures
• Exceed iteration limit without human check

ALWAYS:

• Review changes before commit
• Document what was fixed
• Check for regressions
• Store results in Memory-MCP

Related Skills

• codex-safe-experiment: Sandbox experimentation
• smart-bug-fix: Systematic debugging
• testing-quality: Test generation
• multi-model-discovery: Find existing fixes

Verification Checklist

• All targeted tests passing
• No new regressions
• Changes reviewed
• Memory-MCP updated
• Documentation complete