AgentSkillsCN

Epic Meta Prompting

史诗级元提示

SKILL.md

Epic Meta-Prompting Skill

Generate research and planning prompts for epic workflows using isolated sub-agents.

Purpose

Use meta-prompting to create research and planning pipelines via /create-prompt and /run-prompt, preventing context pollution through isolated execution.

When to Invoke

  • •During /epic Step 3 (Research & Planning phase)
  • •After epic specification is complete and reviewed
  • •Skip if resuming past the planning phase

Research Prompt Generation

bash
/create-prompt "Research technical approach for: [epic objective]"

The create-prompt skill will:

  1. •Detect purpose: Research
  2. •Ask contextual questions (depth, sources, output format)
  3. •Generate research prompt in .prompts/001-[epic-slug]-research/
  4. •Reference relevant project docs (@docs/project/tech-stack.md)
  5. •Specify XML output format with metadata

Auto-execute research prompt:

bash
/run-prompt 001-[epic-slug]-research

Research outputs:

  • •.prompts/001-[epic-slug]-research/research.md
  • •Contains: findings, confidence levels, dependencies, open questions, assumptions

Copy and Commit Research

bash
# Copy research.md to epic directory
cp .prompts/001-${EPIC_SLUG}-research/research.md epics/${EPIC_SLUG}/research.md

# Extract metadata for commit message
FINDINGS_COUNT=$(xmllint --xpath 'count(//finding)' epics/${EPIC_SLUG}/research.md)
CONFIDENCE=$(xmllint --xpath 'string(//confidence_level)' epics/${EPIC_SLUG}/research.md)
OPEN_QUESTIONS=$(xmllint --xpath 'count(//open_question)' epics/${EPIC_SLUG}/research.md)

# Commit research
git add .prompts/001-${EPIC_SLUG}-research/
git add epics/${EPIC_SLUG}/research.md
git commit -m "docs(epic-research): complete technical research for ${EPIC_SLUG}

Findings: ${FINDINGS_COUNT}
Confidence: ${CONFIDENCE}
Open questions: ${OPEN_QUESTIONS}

Next: Planning phase"

Alternative: /meta:enforce-git-commits --phase "epic-research"

Planning Prompt Generation

bash
/create-prompt "Create implementation plan based on research"

The create-prompt skill will:

  1. •Detect purpose: Plan
  2. •Reference research.md from step above
  3. •Generate plan prompt in .prompts/002-[epic-slug]-plan/
  4. •Specify plan.md output with phases, tasks, constraints

Auto-execute plan prompt:

bash
/run-prompt 002-[epic-slug]-plan

Plan outputs:

  • •.prompts/002-[epic-slug]-plan/plan.md
  • •Contains: architecture decisions, implementation phases, dependencies, risks

Copy and Commit Plan

bash
# Copy plan.md to epic directory
cp .prompts/002-${EPIC_SLUG}-plan/plan.md epics/${EPIC_SLUG}/plan.md

# Extract metadata for commit message
ARCH_DECISIONS=$(xmllint --xpath 'count(//architecture_decision)' epics/${EPIC_SLUG}/plan.md)
PHASE_COUNT=$(xmllint --xpath 'count(//phase)' epics/${EPIC_SLUG}/plan.md)
DEPENDENCY_COUNT=$(xmllint --xpath 'count(//dependency)' epics/${EPIC_SLUG}/plan.md)
RISK_COUNT=$(xmllint --xpath 'count(//risk)' epics/${EPIC_SLUG}/plan.md)

# Commit plan
git add .prompts/002-${EPIC_SLUG}-plan/
git add epics/${EPIC_SLUG}/plan.md
git commit -m "docs(epic-plan): create implementation plan for ${EPIC_SLUG}

Architecture decisions: ${ARCH_DECISIONS}
Phases: ${PHASE_COUNT}
Dependencies: ${DEPENDENCY_COUNT}
Risks: ${RISK_COUNT}

Next: Sprint breakdown"

Alternative: /meta:enforce-git-commits --phase "epic-plan"

Auto-Mode Handling

If auto-mode enabled:

  • •Skip PAUSE points (spec review, plan review)
  • •Auto-execute /run-prompt after /create-prompt
  • •Update phases status automatically in state.yaml

If interactive mode:

  • •PAUSE: "Plan review complete. Continue to sprint breakdown? (y/n)"
  • •Update state.yaml with approval status

State Updates

After plan review (interactive):

yaml
manual_gates:
  plan_review:
    status: approved
    approved_at: { ISO_TIMESTAMP }
    approved_by: user

phases:
  research:
    status: completed
    completed_at: { ISO_TIMESTAMP }
  planning:
    status: completed
    completed_at: { ISO_TIMESTAMP }

Velocity Benefit

  • •Research phase: 1-2 hours (vs 4+ hours manual research)
  • •Planning phase: 1-2 hours (vs 3+ hours manual planning)
  • •Total: 3-4x faster with meta-prompting