菜单
For substantial or autonomous multi-step work, commit to a reviewable contract -- intent, a plan with acceptance criteria, then execute proving each unit with evidence -- instead of improvising. Use for overnight/large tasks or anything that must be auditable.
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
When the user asks what is going on with the site they are building, drive the in-app browser Tools endpoints to start the dev server, read console + server + network logs, inspect/eval the page, and report -- all inside Symphonee, no Chrome needed.
The disciplined way to make and land a code change in a Symphonee-managed repo -- branch, implement, verify, show the diff as its own step, then commit. Use for any implement/fix/refactor task that ends in a commit.
How to safely edit and validate Symphonee's dashboard renderer. The served js/app.js and mind-ui.js are GENERATED build outputs -- edit the source under dashboard/public/app/src/shell/ or mind-ui/src/, rebuild, then validate.
Turn a correction, a gotcha, or a procedure you just did well into a durable SKILL.md so every future session of every CLI inherits it. This is how the system improves its own behaviour, not just its knowledge.
Engineered, not rushed -- profile and understand the system before changing it, find root cause not symptoms, then implement deliberately. Use for any non-trivial fix or feature where the right design is not obvious.
When a task matches an installed plugin or a structured connector, use its REST routes -- never re-solve it by hand or shell out to a third-party CLI. Ask the user before invoking a plugin.
| name | Run a contract (intentional autonomous work) |
| description | For substantial or autonomous multi-step work, commit to a reviewable contract -- intent, a plan with acceptance criteria, then execute proving each unit with evidence -- instead of improvising. Use for overnight/large tasks or anything that must be auditable. |
| when | a large, multi-step, or autonomous task where the user wants the result to be trustworthy and provable |
| tags | workflow, contract, autonomy, core |
Turns "improvise a big task" into "commit to a plan and prove you met it." This is how autonomous and overnight work earns trust: the user can see the intent, the acceptance criteria, and the evidence for each unit -- not just a final claim.
POST /api/contracts { title, createdBy, intent: { restatement, constraints[], assumptions[], unknowns[] } }. Restate
the task in your own words; list the real constraints (e.g. "don't push/merge"),
the assumptions you are making, and the unknowns you cannot resolve.POST /api/contracts/update { id, phase:'plan', plan: [{ id, goal, outputs, acceptance, deps }] } -- where every unit has an
explicit acceptance criterion (how you will know it is done) and its deps.review.planApproved. If running autonomously with standing authorization,
record your own plan-review note and proceed.POST /api/contracts/update { id, units:[{ id, status:'done', evidence:'<what proves the acceptance criterion>' }] }. Use the ship-a-change
skill for any code unit.review.finalApproved, and the contract auto-moves to
done. Summarize what was built and what (if anything) was deferred.done without real evidence that its acceptance criterion is
met. No hedging, no "should work".done unit.done.