Free SKILL.md scraped from GitHub. Clone the repo or copy the file directly into your Claude Code skills directory.
npx versuz@latest install joshua-eisenhart-codex-ratchet-system-v4-skill-specs-skill-improver-operatorgit clone https://github.com/Joshua-Eisenhart/Codex-Ratchet.gitcp Codex-Ratchet/SKILL.MD ~/.claude/skills/joshua-eisenhart-codex-ratchet-system-v4-skill-specs-skill-improver-operator/SKILL.md--- skill_id: skill-improver-operator name: skill-improver-operator description: Validate and optionally commit a proposed target-skill update under explicit dry-run, test, and allowlist gates. skill_type: operator source_type: repo_skill applicable_layers: [A2_MID_REFINEMENT, A2_LOW_CONTROL] applicable_graphs: [runtime, a2_mid_refinement_graph_v1] inputs: [target_skill_path, candidate_code, test_command, allow_write, approval_token, allowed_targets, recorder, context_packet, context_bridge_packet, bridge_packet, directive_topics, selected_witness_indices, context_notes] outputs: [improved, detail, path, proposed_change, compile_ok, tests_state, score, write_permitted, dry_run, context_contract_status, context_summary] related_skills: [a2-skill-improver-readiness-operator, autoresearch-operator, llm-council-operator] capabilities: can_write_repo: true requires_human_gate: true tool_dependencies: [] provenance: "Karpathy mutator-loop pattern retooled into a bounded Ratchet skill-validation and gated-commit operator" adapters: codex: system_v4/skill_specs/skill-improver-operator/SKILL.md shell: system_v4/skills/skill_improver_operator.py --- # Skill Improver Operator Use this skill when a candidate update to one existing skill needs bounded validation before any repo write is permitted. ## Purpose - validate one proposed target-skill update - require explicit dry-run/test/allowlist gates - allow commit only when approval is explicit ## Execute Now 1. Read: - [SKILL_IMPROVER_READINESS_REPORT__CURRENT__v1.json](/home/ratchet/Desktop/Codex%20Ratchet/system_v4/a2_state/audit_logs/SKILL_IMPROVER_READINESS_REPORT__CURRENT__v1.json) - [a2-skill-improver-readiness-operator spec](/home/ratchet/Desktop/Codex%20Ratchet/system_v4/skill_specs/a2-skill-improver-readiness-operator/SKILL.md) 2. Require: - `candidate_code` - bounded `test_command` - explicit `allow_write` - explicit `approval_token` - exact-path `allowed_targets` 3. Keep dry-run as the default. 4. Treat any `context_packet` / `bridge_packet` inputs as bounded first-target metadata only. ## Context Contract - Optional context fields may be supplied for the current proven first-target class only: - `context_packet` - `context_bridge_packet` - `bridge_packet` - `directive_topics` - `selected_witness_indices` - `context_notes` - This contract is metadata-only: - it may shape repo-held audit context - it must not change write permission, allowlist gates, compile gates, test gates, or target admissibility - Do not treat this contract as permission for: - second-target admission - live learning - runtime import - graph backfill ## Quality Gates - Do not invent mutations implicitly. - Do not write to a target that is not exact-path allowlisted. - Do not commit a candidate that fails `py_compile`. - Do not commit a candidate when the selected test command fails. - Do not treat this operator as permissionless self-mutation. - Do not treat accepted context metadata as proof that a broader consumer lane is runtime-live.