SKILL.md Template

Specialized instruction sets the agent loads when relevant. Sit at .agents/<mode>/skills/<skill-name>/SKILL.md.

Frontmatter

---
name: rug-check-deep
description: |
  Use when user asks "is this token a rug", "vet this contract",
  "check this Bags.fm token for safety", or any safety/rug evaluation
  question. Runs a 4-stage rug check.
---

The description is what the runtime keyword-matches against. Be specific — include the trigger phrases users say.

Body structure

  1. Step-by-step flow — numbered, concrete
  2. Output format — example showing the response shape
  3. Anti-patterns — what to avoid
  4. Closing note — disclaimers ("not financial advice")

Example

# Deep Rug Check

Run when a user asks if a Bags.fm token is safe.

## Steps

1. `bags_token_score` → composite score
2. `bags_holder_distribution` → top 10 holders %
3. `bags_lp_status` → LP locked? Until when?
4. `bags_mint_authority` → mint/freeze authority status

## Output

```
🎒 Rug check: <token>

Score:    <0-100>
Holders:  Top 10 = <X>%
LP:       <locked until DATE | UNLOCKED ⚠>
Mint:     <revoked | active ⚠>

Verdict: <SAFE | CAUTION | RUG>
```

## Anti-patterns

- Don't tell user it's "safe" — only describe risk factors
- Don't predict price
- Always: "Not financial advice"

Scaffolding

Use the skill-workshop agent to scaffold new skills:

vesper agent run skill-workshop "Write a skill for X"

What's next?