AgentSkillsCN

run-plan

按照技能序列，以确定性的方式执行连续计划。在连续规划器生成包含技能序列的计划后，可使用此技能。它会逐条读取 task_plan.md 文件，并按步骤执行每项技能。

SKILL.md
--- frontmatter
name: run-plan
description: "Execute a continuous plan deterministically by following the skill sequence. Use after continuous-planner creates a plan with skill sequences. Reads task_plan.md and executes each skill step by step."
disable-model-invocation: true
allowed-tools: Read, Write, Edit, Bash, Skill

Run Plan - Deterministic Skill Execution

Execute continuous plans by following the skill sequence step-by-step.

When To Use

After creating a continuous plan with continuous-planner, run this to execute the skill sequence deterministically.

How It Works

  1. •Read task_plan.md to get the skill sequence
  2. •For each skill in sequence:
    • •Check if status is "⏸️ Pending"
    • •If pending, execute the skill
    • •Update status in task_plan.md
    • •Update progress.md with results
  3. •Handle failures with recovery protocols
  4. •Mark plan complete when all skills executed

Skill Sequence Format

The task_plan.md should have a Skill Sequence table:

markdown
| Step | Skill | Purpose | Input | Output | Status |
|------|-------|---------|-------|--------|--------|
| 1 | front-door | Interview/triage | User request | requirements | ⏸️ Pending |
| 2 | api-designer | Design API | requirements | api_spec | ⏸️ Pending |
| 3 | implement-plan | Execute | api_spec | Complete | ⏸️ Pending |

Execution Flow

code
┌─────────────────────────────────────────┐
│  1. Read task_plan.md                   │
│  2. Parse Skill Sequence table          │
│  3. For each pending skill:             │
│     a. Invoke skill with input          │
│     b. Capture output                   │
│     c. Update findings.md               │
│     d. Update progress.md               │
│     e. Update skill status in plan      │
│  4. Update beads                       │
│  5. Report completion                   │
└─────────────────────────────────────────┘

Error Handling

If a skill fails:

  1. •Log error to findings.md
  2. •Check failure-recovery skill for protocols
  3. •Update skill status to "❌ Failed"
  4. •Ask user for guidance

Beads Integration

Each skill execution is tracked as a bead task:

  • •Plan epic: continuous-[project-name]
  • •Skill tasks: Created as subtasks
  • •Completion: Mark beads as complete

Example Usage

bash
# From project directory with continuous plan
/run-plan

# Or specify plan location
/run-plan .claude/continuous/task_plan.md

Output

After execution:

  • •task_plan.md: All skills marked complete or failed
  • •findings.md: Research, errors, discoveries
  • •progress.md: Execution log with checkpoints
  • •beads: Tasks marked complete

Keywords

run plan, execute plan, deterministic, skill sequence, continuous plan, task_plan.md