Use for planning, roadmaps, design, architecture, sequencing, or before consequential work when no current FINAL plan plus GO covers it. Fire before changes to data or schemas, credentials or auth, production or releases, repository or source-home boundaries, irreversible work, or materially costly mistakes. Re-enter when that scope appears mid-task or when restoring prior plan state. Skip factual answers, trivial reversible one-step work, and authorized Builder or Reviewer execution under a current FINAL blueprint plus GO unless a reopen condition fires.
中文简介正在准备中。当前显示原文说明。
实测评分 未实测
查看安装方法安装前可以查看交给 AI 代理的指令内容。
Keep a compact Plan Capsule, not the planning transcript, for material work.
PLAN <id>: DRAFT | CHALLENGE-READY | FINAL
AS OF <latest operator instruction accounted for>
DEPTH: LIGHT | STANDARD | DEEP
ASKED
<operator's words that define the outcome>
GOAL
<accepted restatement>
DELIVERABLE
<what will exist>
PLAN
1. <step with a checkable completion condition>
2. <step with a checkable completion condition>
DONE
<smallest observable acceptance condition>
Add SOURCE HOME, BOUNDARIES, or OPEN only when they carry load-bearing information.
Copy ASKED from the operator. A continuation (proceed, yes) keeps ASKED and advances AS OF. A changed outcome replaces ASKED, updates the plan, and is read back.
Show the full current capsule when a load-bearing part of the plan changes, before challenge, on restore, and before final closeout. Append only applicable temporary sections:
CHANGED
- <changed plan state>
ADDED
- <new agent-added scope or requirement>
DROPPED
- <operator-requested item that disappeared or changed meaning>
Do not persist those temporary sections after the operator has seen them.
LIGHT: bounded, familiar, reversible work. No challenge by default.STANDARD: several meaningful decisions or moderate uncertainty. Challenge when a load-bearing question remains, the operator asks, or error cost warrants it.DEEP: architecture, broad blast radius, authority, safety, recovery, migration, expensive error, or major uncertainty. Challenge by default unless the operator skips it.Raise depth when stakes rise. Read back every depth change.
Before continuing restored planning or execution:
AS OF includes the current operator instruction.If the capsule is complete, exact, and current, continue after the readback. Do not ask for another confirmation only because a session boundary occurred.
If state is incomplete, stale, conflicting, uncertain, or cannot reproduce the complete capsule exactly, mark RESTORED: UNVERIFIED, show the reconstruction, and hold for operator confirmation. A FINAL blueprint plus GO is not an exact restore when the complete capsule cannot be reproduced.
Ask: Is this the plan?
When the operator settles the plan, respond exactly:
Plan is FINAL. Execution waits for your GO.
FINAL never authorizes execution. Execute only after a later, separate, explicit GO.
When a FINAL plan is handed to a Builder or Reviewer, read references/blueprint.md. Reopen planning before changing the objective, deliverable, source home, load-bearing boundary, acceptance condition, or plan currency.
还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Evaluate minimum-sufficient implementation routes and enforce the selected route for a fixed, authorized outcome. Use after outcome, scope, authority, and acceptance evidence are fixed. Do not choose outcomes, set acceptance, adjudicate governance, grade finished work, design tests alone, conduct security audits, or perform unrelated cleanup.
中文简介正在准备中。当前显示原文说明。
Oracle-frozen Builder and independent-Judge convergence cycle. Load ONLY when the operator explicitly sets Governance Dial G2 for the task, or explicitly names this skill or an active work unit already running it. Never auto-activate on task class, such as security, auth, or payments. If a task seems to warrant G2 and the operator has not said so, ask first. Not for ordinary governed implementation, analysis, review-only, or documentation work.
中文简介正在准备中。当前显示原文说明。
Full governance constitution for work that touches shared state, requires independent review, spans seats, or involves planning, architecture, code, reviews, dispatches, status reports, or canonical records. Not needed for quick factual questions or trivial one-off answers.
中文简介正在准备中。当前显示原文说明。
Requires behavioral, failure-path, and durable-seam evidence for tests and review. Use when writing tests, reviewing test coverage, assessing behavioral test quality, or accepting high-risk behavior on test evidence.
中文简介正在准备中。当前显示原文说明。
The working method for every nontrivial task: frame the objective, ground load-bearing facts, converge on a route, execute without drift, and verify before delivery. Use for analysis, planning, implementation, debugging, document work, reviews, and multi-stage tasks.
中文简介正在准备中。当前显示原文说明。