---
name: discover
description: "Recommend the right skills or pipeline given a described goal. Use when you are not sure which skill to reach for, or want to know what combination of skills fits a workflow. Keywords: recommend, find skill, which skill, what skill, help, navigate, discover, pipeline."
argument-hint: "<described goal or workflow>"
allowed-tools: Read, Glob
enabled: false
source: github:supertyrelle/pelley
imported-from: github:supertyrelle/pelley
---

# Discover: Goal-Based Skill Recommendation

You are running the **discover** skill — recommending the right skills or pipeline for a described goal. Goal: **$ARGUMENTS**

## When to Use

- When you are not sure which skill to reach for for a given workflow
- When you want to know which skills compose well for a multi-step pipeline
- When onboarding to the skill system and want an entry point
- When a user describes a goal in natural language and needs a skill recommendation

## Overview

```
Read all skill descriptions
  -> Match goal to 2-4 most relevant skills
    -> Identify canonical pipeline if applicable
      -> Output pipe-format recommendation
```

---

## Phase 1: Load Skill Descriptions

Use Glob to find all skill definition files:

```
skills/*/SKILL.md
```

For each file found, Read it and extract:
- The `name` field from YAML frontmatter
- The `description` field from YAML frontmatter

Do not read the full body of each skill — frontmatter only is sufficient for matching. If a file has no `description` field, skip it.

---

## Phase 2: Match Goal to Skills

Semantically match **$ARGUMENTS** against the collected descriptions.

Select the **2 to 4 skills** that best fit the described goal. Prefer skills where the description's "when to use" language closely matches the goal's framing.

For each selected skill, note:
- **Skill name** (the slash command, e.g., `/gather`)
- **One-line description** drawn from the `description` field
- **Why it fits** — one sentence specific to the user's stated goal

### Canonical Pipeline Detection

Check whether the goal maps to a known pipeline pattern. Known patterns (from `docs/INDEX.md`):

| Goal Type | Canonical Pipeline |
|-----------|-------------------|
| Research a topic | `/gather` -> `/distill` -> `/rank` |
| Compare two approaches | `/gather` -> `/diff-ideas` |
| Plan before building | `/decompose` -> `/plan` -> `/spec` |
| Assess risk before a change | `/gather` -> `/critique` -> `/rank` |
| Explore unknown territory | `/blossom` or `/fractal` |
| Review code before merging | `/review` |
| Run a session | `/status` -> work -> `/retro` -> `/handoff` |
| Manage a team | `/assemble` -> `/standup` -> `/sprint` |
| Train an agent | `/diagnose-agent` -> `/challenge-gen` -> `/challenge-run` |
| Prototype something | `/gather` -> `/distill` -> `/sketch` |
| Build a feature iteratively | `/tracer` |
| Define a testing approach | `/test-strategy` |
| Identify failure modes before building | `/premortem` |
| File a structured bug report | `/review` -> `/bug` |
| Bootstrap a new project | `/bootstrap` |
| Plan goals with a persistent team | `/team-meeting` -> `/sprint` |
| Run full agent improvement cycle | `/diagnose-agent` -> `/challenge-gen` -> `/challenge-run` -> `/active-learn` |
| Facilitate a requirements discussion | `/meeting` |

If the goal maps to a pattern above, include the full pipeline in the output.

---

## Phase 3: Emit Recommendation

Output in pipe format:

```markdown
## Skill Recommendations for: [goal summary]

**Source**: /discover
**Input**: [user's goal, one line]
**Pipeline**: (none — working from direct input)

### Items (N)

1. **/skill-name** — [one-line description from frontmatter]
   - Why it fits: [one sentence specific to the user's goal]

2. **/skill-name** — [one-line description from frontmatter]
   - Why it fits: [one sentence specific to the user's goal]

[... up to 4 items]

### Suggested Pipeline (if applicable)

If your goal maps to a canonical workflow:

`/skill-a` -> `/skill-b` -> `/skill-c`

[One sentence explaining the flow and what each step contributes.]

### Summary

[One paragraph. Recommend where to start, explain why the top skill fits best, and note any tradeoffs between options.]
```

---

## Guidelines

- **Prefer specificity over completeness.** Two well-matched recommendations beat five loosely-matched ones.
- **Name the tradeoff.** If two skills both fit, say which is better for breadth vs. depth, or exploration vs. execution.
- **Do not hardcode the skill list.** Always read from `skills/*/SKILL.md` at runtime so the recommendation reflects the actual installed skill set.
- **Pipeline is optional.** Only include the Suggested Pipeline section if the goal clearly maps to a multi-step workflow. Single-skill goals do not need a pipeline.
- **Argument handling**: If `$ARGUMENTS` is empty, ask the user to describe their goal in one sentence before proceeding.