菜单
Route repository work through the sentropic harness method. Use in a sentropic repo for branch scope, brainstorming, planning, debugging, testing, review, or adopting harness conventions in a new repo.
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。
| name | harness |
| description | Route repository work through the sentropic harness method. Use in a sentropic repo for branch scope, brainstorming, planning, debugging, testing, review, or adopting harness conventions in a new repo. |
Read using-harness/SKILL.md first in a sentropic repository. Then read only the workflow needed for the current task:
brainstorm/SKILL.md before creative or design work.plan/SKILL.md to turn accepted decisions into an executable BRANCH.md.debug/SKILL.md before diagnosing a defect or unexpected behavior.test/SKILL.md when adding or changing tests.review/SKILL.md before completion or merge.adopt/SKILL.md when bringing the method to a new repository.Follow the selected workflow completely. Repository-local AGENTS.md and explicit user instructions take precedence.
Load FIRST in any sentropic repo — establishes the harness method (branch/scope/test/debug/review/brainstorm/plan) and replaces legacy superpowers for those acts.
Use when the user asks for a track report, status, or advancement/progress report — run the CLI `track report` from the repo root and return its AI-prepared, cited output verbatim; never substitute the deterministic `track_report` MCP projection for a human AI report. Also use when an agent needs to read, update, import, or verify track state; when a BRANCH.md or plan/NN-BRANCH_*.md changed; or when deciding between Track MCP and CLI. MCP remains read-only, writes/imports use the CLI, and .track is append-only/single-writer.
Coordinate with other CLI agents (Claude Code, Codex, Gemini) via the h2a protocol — open a live session, list peers, exchange messages, drive a signed negotiation or objective loop. Use when the user wants the current CLI to interact with another agent through a shared workspace.
Use at a branch merge or branch-close to protect track's event log from merge-time loss and to refresh acceptance. DETECT-and-GUIDE only, never auto-repair: judge git ancestry in the shell, detect STRUCTURAL event loss via `track events-contains` (NOT audit.orphan, which is blind to a reparent lost toward a valid parent), recover OPPORTUNISTICALLY from a surviving ref/reflog before declaring anything irrecoverable, then re-anchor acceptance freshness with query/report --require-accepted + consolidate and RE-READ to surface done-but-skipped items. Orchestrates the existing CLI; reimplements nothing.
Use when an agent must present a complex, irreversible, cross-owner, security/contract-affecting, costly, or genuinely balanced decision to a human owner — or must choose between decide-and-trace, a quick ask, and a full decision dossier. The agent presents at the level of the stakes, never sells its preferred option, and records the decision in track. This is agent→human; it is NOT h2a (h2a ATTENTION is the inter-agent presenter).
Use when a flat track backlog has grown past a handful of items and needs structuring into a small set of perennial workpackages (WPs) — durable themes by owning concern, not milestones — so a conductor can pilot %-by-WP. The skill reads the backlog, clusters by durable concern, and EMITS A PROPOSAL (the WP set + a per-item reparent plan) the human ratifies before anything is written. It never auto-restructures; for a consequential restructuring it composes present-decision. This is agent→human structuring of one repo's own backlog.