Skill v1.0.1
currentAutomated scan100/100+2 new
version: "1.0.1"
REQUIRED. Lowercase, hyphens. MUST match the folder name.
name: template-skill
REQUIRED. ≤60 chars, one sentence, ends with a period.
State the capability, not the implementation. No marketing words.
Do NOT repeat the skill name in the description.
description: Replace with the capability this skill provides.
REQUIRED. core | practical | optional
core → always-on discipline (skills/)
practical → task-specific, loaded on trigger (skills/)
optional → heavy/niche, installed explicitly (optional-skills/)
tier: practical
REQUIRED. See spec §1.2 for the valid set.
category: workflow
REQUIRED. human | agent (agent-authored skills are curator-eligible)
created_by: human
REQUIRED. OSes the scripts and shell snippets actually support.
verify.sh audits this against script imports.
platforms: [windows, macos, linux]
Optional. Free-form keywords for search and grouping.
tags: []
Optional but expected. Human contributor first; format: Name (@handle).
author: Your Name (@your-handle)
Optional. Config keys this skill expects under .dojo/config.yaml.
Each key is prompted during dojo setup if missing at load time.
config:
- dojo.example.threshold
Optional. MCP server dependencies. IDs MUST resolve to mcp/registry.yaml.
mcp:
required: [github]
optional: [fetch]
Template Skill
Two to three sentences stating what this skill does AND what it deliberately does NOT do. The "does not" half is as important as the "does" half — it prevents over-triggering on adjacent tasks.
When to Use
Specific triggers, keywords, and contexts that should activate the skill. Be slightly pushy — agents undertrigger more often than they overtrigger.
- Trigger condition 1 (concrete situation, not abstract)
- Trigger condition 2
- Trigger condition 3 (include negative triggers if helpful: "but NOT when…")
Prerequisites
What must be true before the skill can run.
- Tool / MCP server availability
- Required env vars or config keys
- Repo state (e.g. clean working tree, on a feature branch)
If a prerequisite is missing, the skill must fail loudly — never degrade silently.
How to Run
One-screen overview of the canonical happy path. If the skill ships a script, show the command:
scripts/example.sh --flag value
Quick Reference
| Action | Command / Tool | Notes | |
|---|---|---|---|
| Common op 1 | view path | Use the Copilot view tool, not shell cat. | |
| Common op 2 | grep pattern | Use the Copilot grep tool, not shell grep/rg. | |
| Common op 3 | scripts/example.sh | Ships with the skill. |
The agent should be able to act from this table alone for routine cases.
Procedure
Step-by-step detailed workflow. This is where the depth lives.
Step 1: Assess
What the agent checks before acting.
Step 2: Act
The actual work. Reference real Copilot tools in backticks (view, edit, grep, glob, powershell, web_fetch, task). Do NOT instruct the agent to call bare shell utilities (cat, sed, find) that have a Copilot wrapper.
Step 3: Confirm
Intermediate verification before moving on.
Pitfalls
Imperative "DO NOT" register of failure modes this skill prevents.
- DO NOT [common mistake 1] — [why it bites].
- DO NOT [common mistake 2] — [why it bites].
- DO NOT [common mistake 3] — [why it bites].
Verification
How the agent (or a reviewer) proves the skill worked.
- [ ] Test or check 1 (with concrete command)
- [ ] Test or check 2
- [ ]
scripts/verify.shpasses