Skill Finder

---
name: skill-finder
description: "Recommend the right Zynkr assistant(s) for a user's task. Interprets intent, matches against the structured assistant index, filters to usable entries, and either recommends or asks a single clarifying question."
category: product
project: skill-finder
platform: claude
status: Done
author: Peter Tu
sheetId: "5.01"
originalName: "行政助理 ─ 搜尋助理索引"
input: "A user inquiry describing a task, problem, or workflow they want help with"
process: "Interpret intent, match against the structured assistant index using functional descriptions, filter to usable entries, decide whether to recommend or ask one clarification"
output: "A concise recommendation of the most relevant assistant(s), or a clear single-question clarification request, or an honest 'no match' response"
synergy: []
handoff: []
house-style: exempt — pure router — selects a child skill, drafts nothing itself

---

# Skill Finder

```bash
npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill skill-finder
```

Help a user pick the right Zynkr assistant for what they're trying to do. This skill reads the structured assistant index, matches the user's stated need against functional descriptions, and either recommends a small set of assistants or asks one targeted clarifying question. Use it as the front door to the Zynkr marketplace when a user describes a task but doesn't know which skill exists for it.

---

## Step 1 — Load the assistant index

Read the canonical taxonomy and index from these references (relative to `{{SKILL_DIR}}`):
- `./references/assistant-index.md` — list of assistants with category, slug, one-line description, IPO summary, and stage status
- `../../../taxonomy.md` (repo root) — category definitions

If the index file is missing or stale, fall back to fetching the current marketplace API:
```
GET https://zynkr.ai/api/skills
```

---

## Step 2 — Interpret intent

Read the user's request carefully and extract:
1. **The task verb** — what action do they want done? (write / analyze / summarize / interview / translate / etc.)
2. **The artifact** — what's the input or output? (article / CV / meeting notes / lecture / business card / etc.)
3. **The domain** — which functional area? (marketing / sales / operations / training / product / engineering / talent / etc.)
4. **The constraint** — language, scale, format, urgency

Do not ask the user yet. Make a best-effort interpretation first.

---

## Step 3 — Match against the index

Score each indexed assistant on:
- **Task match** (does its `process` cover the user's task verb?)
- **Artifact match** (does its `input`/`output` cover the user's artifact?)
- **Category match** (is it in the right functional domain?)
- **Stage** (Done > WIP > Not started — never recommend a non-Done skill without flagging the status)

Filter to Done assistants unless the user explicitly wants experimental ones.

---

## Step 4 — Decide: recommend, clarify, or decline

Apply these rules in order:

1. **Strong match (1–2 candidates)** → Recommend directly. Show:
   - Skill name + zh-TW alias
   - One-line description
   - Why it fits this user's task (1 sentence)
   - Install snippet
2. **Multiple plausible candidates (3+)** → Ask **one** clarifying question that disambiguates. Examples:
   - "Are you preparing for an interview, or running an interview as the interviewer?"
   - "Is this for a written article, a video, or a lecture?"
3. **No usable match** → Be honest:
   - State that no current assistant matches the task
   - Suggest the closest neighbor (with caveats)
   - Point at the proposal repo: `https://github.com/peter-tu-zynkr/zynkr-skill-idea` so the user can request a new one

Never invent an assistant that doesn't exist in the index.

---

## Step 5 — Output format

For a recommendation:

```
Recommended: <skill-name> (<zh-TW name>)
What it does: <one-line description>
Why it fits: <one sentence connecting user's task to the skill's IPO>
Install: npx skills add https://github.com/peter-tu-zynkr/zynkr-skill-builder --skill <skill-name>
```

For a clarification:

```
One quick question before I recommend:
<single targeted disambiguation question>
```

For no match:

```
I don't see a current Zynkr assistant that matches this. The closest is <skill> but it differs in <X>. If you'd like to propose a new skill, open an issue at https://github.com/peter-tu-zynkr/zynkr-skill-idea.
```

---

## Rules

- One clarifying question max per turn
- Never recommend more than 2 assistants in a single response (forces the user to triage)
- Always include the install snippet so the recommendation is immediately actionable
- Never invent assistant names; only recommend what's in the index
- If the user asks "what assistants are there?" without a task, point them at the index file directly instead of dumping the full list