# H4tcher Adapter Author

> Create or modify a platform adapter with web research, implementation, testing, and D09 checklist verification.

- **Type:** Skill
- **Install:** `agentstack add skill-hatch3r-hatch3r-h4tcher-adapter-author`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [hatch3r](https://agentstack.voostack.com/s/hatch3r)
- **Installs:** 0
- **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [hatch3r](https://github.com/hatch3r)
- **Source:** https://github.com/hatch3r/hatch3r/tree/main/.claude/skills/h4tcher-adapter-author
- **Website:** https://docs.hatch3r.com

## Install

```sh
agentstack add skill-hatch3r-hatch3r-h4tcher-adapter-author
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

> Last updated: 2026-06-09

# Adapter Author

Create or modify a platform adapter for hatch3r.

## Step 0: P8 B1 Ambiguity Gate

Before any read or write, scan the user request against the four B1 triggers from `.claude/rules/clarification-default.md` (which carries the verbatim directive from `governance/CONSTITUTION.md` §2 P8 B1):

1. **Ambiguous scope** — request maps to ≥2 reasonable interpretations producing different adapters (e.g., "add Zed adapter" without specifying rules-only vs full-feature surface).
2. **Multiple valid interpretations** — ≥2 viable approaches with materially different cost, scope, or risk (e.g., bundling MCP support vs. deferring to a follow-up).
3. **Irreversible action** — deleting an existing adapter, renaming its public id, dropping output paths users may already depend on.
4. **Missing acceptance criteria** — no testable definition of done (e.g., no list of file paths, no capability matrix targets, no test coverage threshold).

If any trigger fires, ask via the platform-native question tool per `agents/shared/user-question-protocol.md`: one question per turn; bundle related sub-questions into a single multiple-choice prompt; 2-4 numbered options with one-line trade-offs; declare the default-if-no-response option. Do NOT proceed to Step 1 until the ambiguity is resolved or the user has confirmed the default.

## Step 1: Understand the Contract

1. Read `src/adapters/base.ts` — the `BaseAdapter` abstract class defines:
   - `name: string` — adapter identifier
   - `warnings: string[]` — user-facing warnings
   - `doGenerate(ctx: AdapterContext): Promise` — core generation logic
   - `getOutputPaths(agentsDir, manifest): Promise` — files this adapter creates
2. Read `src/adapters/canonical.ts` — `readCanonicalFiles(canonicalRoot, type, warnings)` reads frontmatter-bearing markdown from the bundled content root resolved by `resolveBundledContentRoot()` (`src/content/contentRoot.ts`), not a user-repo `.agents/` mirror (the legacy `.agents/` materialisation no longer exists; `canonical.ts:1020-1024`)
3. Read `src/adapters/customization.ts` — `applyCustomization()` applies user overrides
4. Read `src/adapters/mcp-utils.ts` — MCP server config transformation

## Step 2: Research the Target Platform

5. Web-search the target platform's official documentation for:
   - Configuration file format and paths
   - Agent/rule/skill capability support
   - MCP server integration format
   - Hook/event system
   - Model configuration syntax
6. Cite documentation version and date in code comments

## Step 3: Study Reference Implementations

7. Read the 3 supported adapters (per CONSTITUTION §6 Decision 12, 1.9.0 hard-cut):
   - `src/adapters/cursor.ts` — full-featured reference (rules, agents, skills, MCP, hooks)
   - `src/adapters/claude.ts` — CLAUDE.md output format
   - `src/adapters/copilot.ts` — GitHub Copilot custom-instructions output
8. Read `governance/audit/domains/D09-platform-adapters.md` for per-adapter audit checklist

## Step 4: Implement

9. Create `src/adapters/{name}.ts`:
   - Export class extending `BaseAdapter`
   - Implement `doGenerate()` — handle each content type (agents, skills, rules, commands, hooks, MCP)
   - Implement `getOutputPaths()` — return all file paths this adapter creates
   - Use `wrapInManagedBlock()` for merge-safe output
   - Use `resolveAgentModel()` for model configuration
10. Register in `src/adapters/index.ts`:
    - Add to `adapterMap`
    - Add to `TOOL_DISPLAY_NAMES`
    - Add to `ADAPTER_CAPABILITIES` matrix

## Step 5: Test

11. Create `src/__tests__/adapters/{name}.test.ts`:
    - Output path correctness
    - Content format validation
    - Feature flag behavior (when features are disabled)
    - MCP format transformation
    - Hook format if supported
    - Managed block markers present
    - Error handling for missing canonical files
12. Run: `npm test`, `npx tsc --noEmit`, `npm run lint`

## Step 6: Verify

13. Run `npx hatch3r validate` with the new adapter configured
14. Verify D09 checklist items: output paths, format, feature flags, bridge orchestration, model emission, MCP, secrets, hooks

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [hatch3r](https://github.com/hatch3r)
- **Source:** [hatch3r/hatch3r](https://github.com/hatch3r/hatch3r)
- **License:** MIT
- **Homepage:** https://docs.hatch3r.com

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-hatch3r-hatch3r-h4tcher-adapter-author
- Seller: https://agentstack.voostack.com/s/hatch3r
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.