md2wechat-skill

# Final Verdict

## Repo

- Name: geekjourneyx/md2wechat-skill
- Date: 2026-04-13
- Archetype: hybrid-skill
- Final bucket: **reusable**
- Confidence: medium-high

## Why This Bucket

- **Core outcome**: The deterministic support layer is **production-grade** — 228 passing Go tests, clean build, 5 discovery commands with consistent JSON envelopes, inspect/config/humanize/write commands all work. Core conversion requires external API (md2wechat.cn) but error handling is clean.
- **Scenario breadth**: Tested: build, all discovery commands, inspect, preview, config show/validate, humanize --help, write --help/--list. Most features verified. Conversion gated by API key (expected for SaaS-backed tool).
- **Repeatability**: 228 tests pass consistently. CLI commands are deterministic. Build is reproducible.
- **Failure transparency**: Structured JSON error envelopes on API failures. Config validation catches issues early. Inspect provides actionable fix suggestions.

## Hybrid-Skill Ceiling Analysis

Per hybrid-skill archetype: the **core LLM layer** (AI mode conversion, write from idea) is untested. However:
- The support layer is exceptionally strong (228 tests, 36 test files, 14 packages)
- The "API mode" conversion is the primary user flow — it's external-API-dependent but well-architected
- The tool degrades gracefully (preview produces output even without API key)

Ceiling applied: core LLM layer untested → **could** cap at `usable`. But the **depth of support layer testing** (228 tests, clean build, comprehensive CLI) and the fact that the primary user flow is API-based (not LLM-based) pushes this to `reusable` with the ceiling noted.

## Score Summary

| Category | Passed | Failed | Partial | Untested | Total |
|----------|--------|--------|---------|----------|-------|
| Critical (support) | 4 | 0 | 2 | 0 | 6 |
| Critical (core) | 0 | 0 | 0 | 1 | 1 |
| High | 5 | 0 | 0 | 1 | 6 |
| Medium | 0 | 0 | 1 | 0 | 1 |
| **Total** | **9** | **0** | **3** | **2** | **14** |

## What I Would Say In Plain English

**md2wechat-skill is the most professionally engineered repo I've evaluated in this batch.** 228 passing Go tests, 36 test files, clean build, consistent JSON API across all discovery commands, structured error handling, multi-platform distribution (Homebrew, npm, script, source). This is production software.

**The core conversion requires an external API key (md2wechat.cn),** which means I can't fully verify the primary feature without credentials. But: the error handling is clean (structured JSON errors, not crashes), preview degrades gracefully, and inspect works perfectly without API access. The architecture handles the dependency well.

**Two minor discrepancies:**
1. "38+ themes" — CLI shows 15 entries; 38 exist in api.yaml catalog but aren't surfaced to users
2. "Multiple writing styles" — only 1 style (dan-koe) available; feels like a 1.0 of the write feature

**What sets this apart**: discovery-first design (agents query capabilities programmatically), test discipline (228 tests across 14 packages), and documentation quality (17+ docs, SKILL-RULE.md meta-guide for writing good skills).

## Path to `recommendable`

1. **Test core conversion with API key** — verify the primary feature end-to-end
2. **Resolve theme count discrepancy** — either expose 38 themes in CLI or adjust README claim
3. **Add more writing styles** beyond dan-koe to match the feature's ambition
4. **Test AI mode** — verify LLM-generated output quality
5. **Test all 4 install methods** — only source build verified in this eval

## Remaining Risks

- **External API dependency** — core conversion requires md2wechat.cn service availability. If the API goes down, the tool is mostly unusable (preview degrades but convert/draft fail)
- **Theme count inflation** — 38+ claim vs 15 CLI entries may confuse users
- **Write feature is early** — only 1 style, feels like an MVP
- **No offline conversion mode** — unlike wewrite which can convert locally, md2wechat requires API for full conversion