AgentSkillsCN

content-pipeline

适用于撰写博客文章、创作内容，或从代码中发掘洞见时使用。触发条件：“generate post”、“create content”、“draft a post”、“find insights”。 采用七阶段结构化内容开发流程，第5阶段之前不进行草稿撰写。

SKILL.md
--- frontmatter
name: content-pipeline
description: |
  Use when generating a post, creating content, or finding insights from code.
  Triggers: "generate post", "create content", "draft a post", "find insights".
  7-phase structured content development. No draft before Phase 5.
user-invocable: true

Content Pipeline

MANDATORY: Follow all 7 phases in order. No draft before Phase 5. No delivery before Phase 6 passes.

Phase 1: Discovery

Read the request carefully. Identify:

  • •Topic area: Which repo(s)? Which product line? (polymaxr, megafi, ai60)
  • •Depth target: Bug story? Architecture decision? Platform quirk? Rewrite story?
  • •Platform: LinkedIn, X/Twitter, or both?
  • •Constraints: Any specific angle the user wants?

If the request is vague ("find something interesting"), scan all three product lines and present options before proceeding.

Phase 2: Exploration

Dispatch the insight-architect agent to explore the target repos. If doing manually:

  1. •Map territory: git log --oneline -30 in target repos. Check branch names for feature work.
  2. •Read gold mines first:
    • •polymaxr/polymaxr-lite/.cursor/rules/common-bugs.mdc
    • •polymaxr/polymaxr-mm-bot/.cursor/rules/common-bugs.mdc
    • •polymaxr/polymaxr-lite-v2/risk/src/ (constants)
    • •megafi/server/src/common/ (infrastructure decisions)
    • •ai60/agentic-org/docs/AGENT-PATTERNS.md
  3. •Parallel mining: When scanning multiple repos, dispatch concurrent reads. Don't scan sequentially when you can scan in parallel.

Do NOT proceed until you have at least 3 candidate insights with source references.

Phase 3: Insight Extraction

For each candidate from Phase 2, apply the depth filter:

  1. •What happened? (1-2 sentences, factual)
  2. •Why did it happen? (first why)
  3. •Why was that the case? (second why, where depth lives)
  4. •What's the structural principle? (transferable lesson)
  5. •What would surprise someone who hasn't built this? (non-obvious part)

If questions 3-5 can't be answered, the insight is too shallow. Discard it.

Present the top 1-3 insights to the user with the depth filter answers. Get approval before drafting.

Phase 4: Design

For the approved insight:

  1. •Choose framework: Apply /copywriting skill to select PAS/AIDA/BAB/4Ps/StoryBrand based on insight type.
  2. •Outline structure: Hook, body sections, closing statement.
  3. •Identify the non-obvious moment: Where in the post does the reader learn something they didn't expect?
  4. •Identify the vulnerable moment: Where do you admit something is hard, broken, or incomplete?

Present the outline to the user. Do NOT proceed to drafting without approval.

Phase 5: Draft

Write the post following:

  • •The approved outline from Phase 4
  • •The chosen copywriting framework (applied invisibly)
  • •All rules from .claude/rules/ (never-do, always-do, depth, voice)
  • •Platform formatting requirements (LinkedIn: 150-300 words; X: 280 chars)

Apply anti-AI constraints WHILE drafting, not after. Build the post clean from the start.

Phase 6: Review

Run three review passes:

  1. •Anti-AI Review: Dispatch anti-ai-reviewer agent. If verdict is CONTAMINATED, fix all findings and re-run.
  2. •Draft Review: Dispatch draft-reviewer agent. If verdict is NEEDS WORK, fix all findings and re-run.
  3. •Scoring Loop: Run /10-10-post skill. If any criterion below 8, rewrite and re-score. Max 5 iterations.
  4. •Humanizer Pass: Run /humanizer skill. Apply the 3-pass system (detect, rewrite, final check).
  5. •Final Verification: Dispatch post-verifier agent. Must return VERIFIED.

Phase 6 is not complete until all five steps pass. If stuck after 3 full cycles, present the best version to the user with the remaining issues flagged.

Phase 7: Completion

Deliver:

  • •Final draft text (ready to copy-paste)
  • •Platform tag (LinkedIn / X / Both)
  • •Source reference (repo, file path or commit SHA)
  • •Content type (bug story / architecture decision / platform quirk / agent lesson / performance story / rewrite story)

Log to memory:

code
[date] | [platform] | [content type] | [source repo] | [topic summary]

Rules

  • •NEVER skip Phase 2 (Exploration). Insights without code exploration are shallow.
  • •NEVER skip Phase 3 (Insight Extraction). The depth filter is mandatory.
  • •NEVER draft before Phase 5. Design first.
  • •NEVER deliver before Phase 6 passes. Review is mandatory.
  • •If scope changes mid-pipeline (user wants a different angle), restart from Phase 3, not Phase 5.
  • •One post per pipeline run. Don't batch.