菜单
HarnessX Critic (Tier 2.1) — adversarial review of evolved-skill proposals against trace evidence. Detects reward hacking and manifest/evidence contradictions. Out-of-band LLM counterpart to the in-loop deterministic critic in src/evolve/critic.rs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | _critic |
| description | HarnessX Critic (Tier 2.1) — adversarial review of evolved-skill proposals against trace evidence. Detects reward hacking and manifest/evidence contradictions. Out-of-band LLM counterpart to the in-loop deterministic critic in src/evolve/critic.rs. |
In-loop vs out-of-band. epic-harness forbids external LLM calls from production, so the reflect loop ships a deterministic critic (
src/evolve/critic.rs) that gates seeding when reward hacking is suspected. THIS skill is the out-of-band LLM version a meta-agent or human runs during/evolvereview for the cases the deterministic check cannot catch (non-local effects, manifest/evidence nuance).
/evolve review of newly seeded skillsreward_hacking_suspected is true in metricsFor each proposal, ask: does the trace evidence support the predicted_impact?
| Excuse | Rebuttal | Do instead |
|---|---|---|
| "The score went up, so it works" | Score can rise via metric gaming | Verify the outcome improved, not just the score |
| "The seesaw passed, it's safe" | Seesaw is coarse; sub-threshold coupling evades it | Check dimension deltas, not just aggregate pass |
| "It's just a prompt tweak" | Prompt edits have non-local effects on shared context | Trace the effect across skills, not just the target |
Ring 3 evolution engine. Analyzes session observations, generates and improves evolved skills, shows metrics dashboard. Subcommands: status, history, rollback, reset. Use for post-session review and skill improvement.
State-persisted autonomous pipeline: spec → go → audit → eval → ship → evolve in one command. Auto-detects direct/council/interactive mode. Crash-recoverable via PIPELINE-*.json. Hands-off until PR.
Core router. Always active. Auto-invokes matching skill before every response. Runs confusion protocol on high-risk ambiguity.
Loop-breaking self-diagnosis. Use when 3+ consecutive failures occur, circular retries persist, or context overwhelms the session.
Audit phase. Parallel review: code quality + security + tests. Semantic dedup of cross-mode findings. Outputs PASS/WARN/FAIL per dimension. Validates spec coverage.
Conventional Commits 1.0 generator. Stages relevant files, infers type(scope): description, never uses --no-verify.