--- name: ES_code-self-review tier: journeyman description: Evolutionary Skill paired with code-self-review. Refines OS through controlled Layer 2 modifications based on execution experience and comprehensive evaluation. version: 0.1.0 author: ChromaCore created: 2026-01-04 binding: os_path: ..\/SKILL.md os_name: code-self-review last_evolution: null --- # ES_code-self-review Evolution system for code-self-review. ## Binding Information **Paired Operational Skill:** code-self-review **OS Path:** ..\/SKILL.md **Binding Type:** 1:1 (this ES exclusively refines the paired OS) ## Evolution Authority This ES has authority to modify: - ✓ Layer 2 content (parameters, wisdoms, heuristics) - ✗ Layer 1 content (skeletal workflow) **Exception cases for L1 modification:** - Fatal flaw causing 100% failure rate - Security vulnerability - Paradigm shift invalidating core assumptions (All require explicit justification in changelog) ## Layer 1\/Layer 2 Boundary Definition ### What is Layer 1 (Immutable) Layer 1 content is wrapped in `<\/L1>` tags throughout the OS. **DO NOT MODIFY Layer 1 content** except in cases of: - Fatal workflow breakage - Security vulnerabilities - Paradigm shifts (requires human approval) ### Layer 1 Content Map The following sections are marked as Layer 1 in code-self-review: - SKILL.md lines 103-111: Core Routing Workflow - SKILL.md lines 114-116: Evolution Check - SKILL.md lines 137-140: Core workflow section **L1 Coverage:** 16\/159 lines (10.1%) ### What is Layer 2 (Tunable) Everything NOT wrapped in L1 tags, including: - Optimization parameters (retry counts, timeouts, thresholds) - Wisdoms section (append-only) - Heuristics and shortcuts - Examples and explanations - Resource file implementations (when not part of L1 workflow) **Layer 2 is your modification target.** ## Evaluation Phase Execute the evaluation script: ```bash python ..\/evolutionary\/scripts\/evaluate_os.py \ --os-path ..\/SKILL.md \ --output eval_report.json ``` Load eval_report.json into context for decision-making. **Evaluation provides:** - Security analysis (5 layers) - Quality metrics (code\/docs\/structure\/functionality) - Compliance validation - Utility assessment - Specific improvement recommendations ## Evolution Workflow ### Phase 1: Context Gathering **Analyze recent OS execution:** - What friction occurred during workflow? - What succeeded vs struggled? - Edge cases discovered? - Performance observations? **Read evaluation report** from previous section. **Review changelog:** ```bash cat ..\/evolutionary\/resources\/os_changelog.md | tail -n 20 ``` **Context to consider:** - Fresh execution experience (just used the skill) - Evaluation metrics (comprehensive analysis) - Evolution history (what's been tried before) ### Phase 2: Generate Modification Candidates Based on all gathered context, propose Layer 2 improvements. **Focus on:** - Tuning parameters (retry counts, timeouts, thresholds) - Adding wisdom entries (tacit knowledge from usage) - Optimizing heuristics (conditional shortcuts) - Refining examples\/documentation (clarity improvements) **For each candidate, include:** - Description of change - Expected impact - Execution cost - Confidence level **Output complete candidate list to lock it in context:** ``` MODIFICATION CANDIDATES: 1. [Candidate description with impact\/cost\/confidence] 2. [Candidate description with impact\/cost\/confidence] ... ``` If no candidates → end evolution workflow. ### Phase 3: Apply 1\/8 Selection Rule Calculate maximum modifications allowed: ``` max_modifications = max(1, ceil(total_candidates \/ 8)) ``` **Score candidates by:** - quality_gain - execution_cost - Confidence level - Alignment with skill's purpose **Output selected candidates to lock them in context:** ``` SELECTED MODIFICATIONS (top [max_modifications]): 1. [Selected candidate with rationale] 2. [Selected candidate with rationale] ... ``` ### Phase 4: Apply Modifications Make the prescribed changes to Layer 2 content: - Update parameter values - Append wisdom entries - Refine heuristics - Improve documentation **CRITICAL:** Do NOT modify anything inside `<\/L1>` tags. ### Phase 5: Post-Modification Steps After applying changes: 1. **Update changelog:** ```bash cat >> ..\/evolutionary\/resources\/os_changelog.md << 'EOF' ## [01052026 0223] Evolution Event **Modifications Applied:** [list changes] **Evaluation Metrics:** [key metrics from evaluation] **Rationale:** [why these changes] EOF ``` 2. **Update timestamp** in OS evolution check step to current time 3. **Increment version** (patch bump) Changes are now committed - proceed to validation. <\/L1> ## Validation Phase Execute the test suite: ```bash python ..\/evolutionary\/scripts\/test_os.py --verbose ``` **If all tests pass:** - Changes are valid - Skill is functional - Evolution successful → commit changes **If any test fails:** - Changes broke the skill - Revert ALL modifications - Log failure in changelog: ```bash cat >> ..\/evolutionary\/resources\/os_changelog.md << 'EOF' ## [01052026 0223] Evolution Attempt Failed **Attempted Modifications:** [list attempted changes] **Validation Failure:** [what broke] EOF ``` - End evolution workflow