Skip to content

chain-authoring-example.md

Latest commit

 

History

History
214 lines (165 loc) · 6.33 KB

chain-authoring-example.md

File metadata and controls

214 lines (165 loc) · 6.33 KB

Chain Authoring: Documentation-to-Skill Pipeline

Build a 4-phase chain that researches library docs and outputs Claude Code skills.

The Request

run >>create_prompt then we will begin crafting a prompt that acts as a
scraper for creating skills... search for the latest information, scrape
docs comprehensively, read backlinks, then create a well-organized skill.md

Natural language → structured chain. Four phases emerged:

Phase Purpose
Discovery Find latest version, map essential docs
Scraping Fetch docs, extract APIs, follow backlinks
Analysis Identify patterns, prioritize content
Synthesis Create token-efficient skill files

Step 1: Create Step Prompts

Each phase becomes its own prompt. Create them first—the chain references them.

prompt_engine(command:">>create_prompt", options:{
  "id": "doc_discovery",
  "name": "Documentation Discovery",
  "category": "knowledge-capture",
  "description": "Find latest version, map essential docs, create prioritized inventory",
  "userMessageTemplate": "# Documentation Discovery: {{library_name}}\n\n**Use Context:** {{use_context}}\n\n## Tasks\n\n### 1. Version Detection\n- Search for latest stable version\n- Identify breaking changes\n\n### 2. Documentation Mapping\n- Main docs URL\n- API reference\n- Migration guides\n\n### 3. Backlinks\n- GitHub repo\n- Changelog\n- Ecosystem integrations",
  "arguments": [
    {"name": "library_name", "type": "string", "description": "Target library"},
    {"name": "use_context", "type": "string", "description": "Primary use case"}
  ]
})

Repeat for doc_scraper, doc_analyzer, skill_synthesizer. Each has focused instructions for its phase.

Step 2: Create the Chain

Reference step prompts in chainSteps. Add gates for quality checkpoints.

prompt_engine(command:">>create_prompt", options:{
  "id": "skill_from_docs",
  "name": "Skill From Documentation",
  "category": "knowledge-capture",
  "description": "4-phase workflow: Discovery → Scraping → Analysis → Synthesis",
  "userMessageTemplate": "# Skill Creation Pipeline: {{library_name}}\n\n| Field | Value |\n|-------|-------|\n| Library | {{library_name}} |\n| Use Context | {{use_context}} |\n| Output Type | {{skill_type}} |\n| Scope | {{scope}} |\n\n## Begin Phase 1\n\nStart with documentation discovery for **{{library_name}}**.\n\nUse:\n- `WebSearch` for official docs and version\n- `context7` MCP if library indexed\n- `WebFetch` to verify URLs",
  "arguments": [
    {"name": "library_name", "type": "string", "description": "Target library"},
    {"name": "use_context", "type": "string", "description": "Primary use case"},
    {"name": "skill_type", "type": "string", "defaultValue": "skill"},
    {"name": "scope", "type": "string", "defaultValue": "global"}
  ],
  "chainSteps": [
    {"promptId": "doc_discovery", "stepName": "Phase 1: Documentation Discovery"},
    {"promptId": "doc_scraper", "stepName": "Phase 2: Comprehensive Scraping"},
    {"promptId": "doc_analyzer", "stepName": "Phase 3: Pattern Analysis"},
    {"promptId": "skill_synthesizer", "stepName": "Phase 4: Skill Synthesis"}
  ],
  "gateConfiguration": {
    "inline_gate_definitions": [
      {
        "name": "Version Verified",
        "description": "Latest version confirmed before scraping",
        "pass_criteria": ["Version number stated", "Release date verified"],
        "apply_to_steps": [1]
      },
      {
        "name": "Token Efficient",
        "description": "Output is dense and actionable",
        "pass_criteria": ["Tables over prose", "No redundant content"],
        "apply_to_steps": [4]
      }
    ]
  }
})

Step 3: Execute

Run the chain with arguments:

prompt_engine(command:">>skill_from_docs library_name=\"Zod\" use_context=\"form validation in Next.js\"")

Phase 1 output:

Chain: chain-skill_from_docs#1
→ Progress 1/4

Execute the phase—research docs, build inventory:

## Version Info
| Field | Value |
|-------|-------|
| Latest Stable | 4.3.5 |
| Release Date | July 2025 |
| Breaking Changes | v3→v4: String validators top-level |

## Priority 1 Docs
| Doc | URL | Purpose |
|-----|-----|---------|
| Official | https://zod.dev | Overview |
| API | https://zod.dev/api | Schema reference |
| Migration | https://zod.dev/v4/changelog | v3→v4 changes |

Pass the gate and continue:

prompt_engine(
  chain_id: "chain-skill_from_docs#1",
  gate_verdict: "GATE_REVIEW: PASS - Version 4.3.5 confirmed, 6 priority docs mapped",
  user_response: "[Phase 1 inventory output]"
)
→ Progress 2/4

Repeat for each phase. Gate verdicts validate quality at checkpoints.

Step 4: Final Output

After Phase 4 synthesis:

→ Progress 4/4
✓ Chain complete

Generated skill structure:

skills/zod/
├── SKILL.md           # Index, quick ref, v4 gotchas
├── api-reference.md   # Primitives, composites, modifiers
└── patterns.md        # Form validation, error handling

SKILL.md excerpt:

# Zod Skill

TypeScript-first schema validation. v4.3.5 (July 2025)

## Quick Reference
| Task | Code |
|------|------|
| Define schema | `z.object({ name: z.string() })` |
| Validate | `schema.parse(data)` or `.safeParse(data)` |
| Get TS type | `type T = z.infer<typeof schema>` |
| Email (v4) | `z.email()` NOT `z.string().email()` |

## v4 Breaking Changes
| v3 | v4 |
|----|-----|
| `z.string().email()` | `z.email()` |
| `{ message: "..." }` | `{ error: "..." }` |

Key Patterns

Pattern Why
Step prompts first Chain references them by promptId
Focused step instructions Each phase has clear scope
Gates at checkpoints Validate before proceeding
user_response carries output Model context has history
gate_verdict advances chain Self-evaluation against criteria

When to Use Chains

Scenario Single Prompt Chain
One-shot task ✓
Multi-phase workflow ✓
Quality gates between phases ✓
Reusable step prompts ✓
Progress visibility ✓

Related