AgentSkillsCN

plan-executor

执行由自主规划器生成的已冻结、经验证的计划，全程零解释、零交互，并提供严格的终止保障。仅当计划已被明确确认、不可更改且执行安全时方可使用。触发关键词包括：“执行计划”“运行执行”“落实已定计划”。

SKILL.md
--- frontmatter
name: plan-executor
description: Executes frozen, validated plans produced by an autonomous planner with zero interpretation, zero interaction, and strict termination guarantees. Use only when a plan is explicitly finalized, immutable, and execution-safe. Trigger keywords: execute plan, run execution, enact finalized plan.
compatibility:
  - requires planner output with explicit FINALIZED, EXECUTION-READY, and IMMUTABLE markers
allowed-tools:
  - system-io
metadata:
  version: 1.0.0
  owner: user

Activation Criteria

Activate this skill if and only if all conditions below are satisfied simultaneously:

  • •A single plan is present.
  • •The plan is explicitly marked FINALIZED, EXECUTION-READY, and IMMUTABLE.
  • •The plan contains a finite, ordered list of steps with contiguous numeric indices starting at 1.
  • •Each step includes:
    • •a single concrete action
    • •a clearly defined target
    • •explicit inputs
    • •an explicit success condition
    • •an explicit failure condition
  • •No step references planning, validation, ideation, optimization, or user feedback.
  • •No step depends on external judgment, preference, or discretion.

Do not activate this skill under any other circumstances.

Execution Steps

  1. •Enter execution mode. From this point, no reinterpretation, planning, validation, or clarification is permitted.
  2. •Lock the plan state. Treat all plan content as read-only.
  3. •Perform preflight verification:
    • •Verify step indices are contiguous and unique.
    • •Verify all required fields are present for every step.
    • •Verify the total number of steps is finite.
    • •Verify no step references undeclared resources or targets.
  4. •If preflight verification fails, halt immediately.
  5. •Execute steps strictly in ascending numerical order.
  6. •For each step:
    • •Execute the action exactly as written.
    • •Apply inputs exactly as specified.
    • •Monitor only the declared success and failure conditions.
    • •Do not retry unless explicitly stated in the step.
  7. •If a step reports success, proceed to the next step.
  8. •If a step reports failure, halt immediately.
  9. •Continue until all steps are completed successfully or execution is halted.
  10. •Exit execution mode.

Ambiguity Handling

  • •Any ambiguity, omission, contradiction, implicit assumption, or multiple interpretations is a fatal error.
  • •Ambiguity includes but is not limited to:
    • •vague verbs
    • •unspecified targets
    • •conditional language without explicit branches
    • •references to “best,” “optimal,” “reasonable,” or similar terms
  • •On detection, halt execution immediately without recovery attempts.

Constraints & Non-Goals

  • •This skill must not plan, replan, revise, optimize, or extend the plan.
  • •This skill must not infer intent, preferences, or context.
  • •This skill must not ask questions or request confirmation.
  • •This skill must not continue after any failure or violation.
  • •This skill must not perform actions outside the explicit scope of the plan.
  • •This skill must not execute indefinitely.
  • •This skill must not output intermediate commentary, logs, or explanations.

Guardrails

  • •Execution scope is strictly limited to declared actions and targets.
  • •No side effects outside declared targets are permitted.
  • •Irreversible actions are prohibited unless explicitly declared and justified in the plan.
  • •Cascading effects not explicitly described invalidate the plan.
  • •Any detected deviation between declared behavior and actual behavior causes immediate halt.
  • •The skill must remain passive unless executing a valid step.

Termination Rules

  • •Normal termination occurs only after the final step completes successfully.
  • •Failure termination occurs immediately on any error, ambiguity, or rule violation.
  • •User-issued stop command causes immediate termination.

Failure Behavior

  • •On failure termination, output a single execution-failure notice stating execution halted due to invalid or unsafe conditions.
  • •On user-issued stop command, output exactly one dot (.).
  • •On successful completion, output nothing.