菜单
Read repo/session state and choose the single next skill: exploit (evolve/polish), explore (pursue/meta-harness/breakout), bootstrap (eval-agent), diagnose (diagnose/eval-harness-diagnose), or step back (reflect). One dispatch, then exit.
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
Raise the ceiling, don't climb toward it. When a target is nearly hit or a metric has plateaued, question the target itself: name the binding constraint, separate the physics floor from the assumed floor, design the regime change that makes 10x reachable, commit through the valley. Triggers: 'breakout', '10x this', 'raise the ceiling', 'why are we capped', 'think way bigger', 'this metric is a cage'.
Prove that a merged change is actually live and production behavior is measured against the deployed artifact. Use after merge, before claiming deployed/live/validated, for Cloudflare Pages/Workers deploys, cache claims, perf claims, and release closeout.
Diagnose and improve eval harnesses built on @tangle-network/agent-eval or similar trace-first systems. Use when success/failure rates, scorecard deltas, judge outputs, promotion gates, benchmark rows, or improvement campaigns may be contaminated by harness bugs, evaluator drift, routing/auth failures, missing traces, or invalid baselines.
Goal-pursuit loop: measure, diagnose, experiment, verify, compare, and iterate against a measurable target. Triggers: evolve, optimize, make this better, push to target.
Split one messy experiment branch into N clean, independent branches — one logical change each, built from the merge-base, disjoint change-sets, each an isolated reviewable PR. Triggers: 'finalize', 'split this branch', 'atomize this', 'make reviewable PRs', 'decompose this mess'.
Adopt @tangle-network/hub-sdk in any product that needs to talk to the Tangle hub (OAuth connections, tools/search/describe/invoke, capability tokens, policies). Use whenever a product imports @tangle-network/agent-integrations, maintains a local hub-client.ts, or hand-rolls fetch() to /v1/hub/*. Forces kill-and-replace, not additive code.
| name | governor |
| description | Read repo/session state and choose the single next skill: exploit (evolve/polish), explore (pursue/meta-harness/breakout), bootstrap (eval-agent), diagnose (diagnose/eval-harness-diagnose), or step back (reflect). One dispatch, then exit. |
Use this only to choose the next skill. It does not perform the work; it dispatches once and exits.
.evolve/ state, recent skill runs, progress, scorecards, and active pursuit files.hypothesize: about to optimize but "what to try" isn't obvious, or the field keeps repeating one idea, or no one surveyed how the world beats this ceiling — build a researched, ranked portfolio before spending compute.evolve: measurable target, metric can still move, and the next bet is clear (or /hypothesize already ranked it).polish: fixed-quality cleanup on existing work.pursue: current approach is wrong or needs a designed generation.meta-harness: architecture evolution can be automated against a benchmark.breakout: near the target or plateaued across 2+ pursue/meta-harness cycles — the ceiling is the target. Question and raise it before climbing again.eval-agent: no evaluator exists for a subjective target.eval-harness-diagnose: an eval harness exists but pass/fail is suspect — deltas can't be attributed, or repeated auth/route/judge/baseline failures suggest harness contamination, not agent failure.diagnose: many failures need clustering and ROI order.finalize: an experiment/pursuit branch carries several mixed changes worth shipping — split it into clean atomic PRs before landing.reflect: the useful work is learning from sessions/projects.stop: no useful next skill or the task is already complete.Return one line: Next: /skill — reason, target, and first check.
If dispatching would be unsafe, return Stop: with the blocking fact.
Use references/full-reference.md for the full decision matrix and state details.
Stop after dispatch; the selected skill owns the next turn.