Skip to main content
Learn by example. This page provides complete skill walkthroughs you can follow to create useful skills for your agents. Each example includes the full SKILL.md file, an explanation of how it works, and how to test it.

​
Built-in skill walkthrough: log-troubleshooting

Comis ships with seven prompt skills in the repo’s /skills/ directory, and the easiest way to understand the format is to read a real one. Here is the manifest and body of log-troubleshooting, the skill agents reach for when the user asks “what went wrong?” or “why is the daemon slow?” Live at: <repo>/skills/log-troubleshooting/SKILL.md 
---
name: log-troubleshooting
description: |
  Investigate and troubleshoot daemon logs at ~/.comis/logs/. Covers
  NDJSON log format, Pino level codes, field dictionary, and staged
  analysis strategies. Use this skill whenever the user asks about logs,
  errors, warnings, daemon issues, slow operations, debugging daemon
  behavior, checking what happened, or investigating any kind of
  platform issue -- even if they don't explicitly mention "logs".
comis:
  requires:
    bins: ["python3"]
---

# Log Troubleshooting

You have read-only access to the daemon logs directory at `~/.comis/logs/`.

This skill bundles `scripts/log-digest.py` -- a Python CLI that parses
NDJSON logs and produces structured, context-friendly summaries...
What this teaches you about SKILL.md authoring:
  • Two required fields, that is it. name and description — the rest is optional. Notice there is no allowedTools, no permissions, no inputSchema. Most real-world skills are this lean.
  • The description does the heavy lifting. Look at the trigger phrases packed into it (“logs”, “errors”, “warnings”, “what happened”, “investigating any kind of platform issue”). Comis uses this string to match the user’s intent to the right skill — a vague description like “log helper” would never get selected. Spend time on this field.
  • comis.requires.bins is how the skill says “I need python3 on PATH”. At load time the runtime eligibility system checks this; if Python is missing, the skill is skipped with a warning rather than silently failing.
  • The body is just Markdown. No special syntax beyond standard {template} / $1 placeholders. The skill bundles a Python script next to the manifest in scripts/log-digest.py, and the body teaches the agent to call it.
Test it (after rebuilding): 
> "Can you check the daemon logs and tell me what happened in the last hour?"
The agent automatically loads log-troubleshooting, runs python3 scripts/log-digest.py ~/.comis/logs/daemon.log, and reports the top errors, slowest operations, and warnings.
Read the rest in <repo>/skills/log-troubleshooting/SKILL.md. The other six shipped skills (chart-visualization, deep-research, find-skills, image-generation, podcast-generation, video-generation) follow the same structure — skim them when picking a template for your own.

​
Customer support persona

A prompt skill that teaches the agent to respond as a customer support representative with specific tone, escalation rules, and response templates. Create the file at: ~/.comis/skills/customer-support/SKILL.md 
---
name: customer-support
description: "Customer support persona with escalation rules"
userInvocable: true
allowedTools: ["message", "memory_search"]
---

You are a customer support representative for our company.

## Tone
- Always be polite, empathetic, and professional
- Use the customer's name when possible
- Acknowledge their frustration before offering solutions

## Escalation Rules
- If the customer asks for a refund, search memory for their order history first
- If no resolution after 3 exchanges, offer to escalate to a human
- Never promise specific timelines without checking

## Response Format
1. Acknowledge the issue
2. Provide a solution or next step
3. Ask if there is anything else you can help with
How it works:
  • The allowedTools restriction ensures the agent only uses messaging and memory tools while this skill is active — it cannot run shell commands, browse the web, or access other tools
  • The userInvocable: true setting (which is the default) means anyone in the chat can activate it with /skill:customer-support
  • The agent follows the tone, escalation, and response format rules for every reply
Test it: 
/skill:customer-support
Then send a message like: “I ordered a product last week and it still has not arrived.”

​
Daily briefing

A prompt skill that generates a morning briefing when triggered by a scheduled cron job. This example shows how skills and scheduling work together. Create the file at: ~/.comis/skills/daily-briefing/SKILL.md 
---
name: daily-briefing
description: "Generate a morning briefing with weather, news, and tasks"
disableModelInvocation: false
argumentHint: "[timezone]"
---

Generate a morning briefing for {timezone}.

Include:
1. A friendly good morning greeting
2. Use web_search to find today's top 3 news headlines
3. Use memory_search to find any pending tasks or reminders
4. Keep it concise -- no more than 10 lines total

Format as a clean, readable message suitable for a chat channel.
How it works:
  • The argumentHint shows users that the skill expects a timezone argument
  • The {timezone} placeholder gets replaced with whatever the user provides
  • The disableModelInvocation: false setting (the default) means the agent can also decide to use this skill on its own when relevant
Test it manually: 
/skill:daily-briefing US/Eastern
Automate it with scheduling: To run this automatically every morning, use the cron tool to create a scheduled job: 
action: add
name: morning-brief
schedule_kind: cron
schedule_expr: "0 9 * * *"
timezone: America/New_York
payload_kind: agent_turn
payload_text: "/skill:daily-briefing US/Eastern"
This tells the agent to invoke the daily briefing skill every day at 9:00 AM Eastern.

​
Code review checklist

A prompt skill with tool restrictions that provides a structured code review checklist. This example demonstrates how to combine instructions with specific tool access. Create the file at: ~/.comis/skills/code-review/SKILL.md 
---
name: code-review
description: "Structured code review with checklist"
argumentHint: "[file-path]"
allowedTools: ["read", "grep", "find"]
comis:
  requires:
    bins: ["git"]
---

Review the code at $1 using this checklist:

## Security
- [ ] No hardcoded secrets or API keys
- [ ] Input validation on all user-facing functions
- [ ] No SQL injection vulnerabilities

## Quality
- [ ] Functions are under 50 lines
- [ ] No duplicated logic
- [ ] Error handling is present (no silent failures)

## Style
- [ ] Consistent naming conventions
- [ ] Comments explain "why", not "what"

Use the read tool to examine the file. Use grep to check for patterns.
Report findings with pass/fail for each item.
How it works:
  • The $1 placeholder is replaced with the file path the user provides
  • The allowedTools restriction limits the agent to read-only file tools — it cannot modify files, run commands, or access the network during the review
  • The comis.requires.bins field ensures git is installed on the system before the skill is loaded. If git is missing, the skill is marked as ineligible.
Test it: 
/skill:code-review src/api/handler.ts
The agent reads the file, checks each item on the checklist, and reports its findings.

​
Data formatter

A skill that formats CSV data into a readable Markdown table. This example shows how permissions control file access. Create the file at: ~/.comis/skills/csv-formatter/SKILL.md 
---
name: csv-formatter
description: "Format CSV data into a readable table"
argumentHint: "[path-to-csv-file]"
permissions:
  fsRead: ["./data/*.csv"]
---

Parse the CSV data provided and format it as a Markdown table.

Input: $1 (path to CSV file or raw CSV text)

Rules:
- Align columns neatly
- Truncate cells longer than 30 characters with "..."
- Show row count at the bottom
How it works:
  • The permissions.fsRead field grants the skill read access to CSV files in the ./data/ directory — and only that directory
  • The skill cannot read files outside of ./data/*.csv, write any files, or access the network
  • The argumentHint shows users what input to provide
Test it: 
/skill:csv-formatter ./data/report.csv
The agent reads the CSV file, parses the contents, and returns a neatly formatted Markdown table.

​
Tips for writing good skills

These tips apply to all prompt skills:
  1. Keep instructions specific and actionable. Instead of “be helpful”, write “respond with a numbered list of 3 options, each with a one-sentence explanation”.
  2. Use numbered steps for procedures. The agent follows numbered steps more reliably than unstructured paragraphs.
  3. Restrict tools to only what the skill needs. Use allowedTools to follow the principle of least privilege. A research skill should not have access to file editing tools.
  4. Test with simple inputs first. Start with the most basic invocation and work up to complex cases. Check that template variables resolve correctly.
  5. Use argumentHint to guide users. Show users what inputs your skill expects so they do not have to guess. For example: argumentHint: "[language] [file-path]".
  6. Write clear descriptions. The agent uses the description field to decide when to apply skills automatically. A vague description like “helper skill” gives the agent little to work with.
Skills are the primary way to customize your agent’s behavior. Start with one or two simple prompt skills, and expand as you learn what works best for your team.

Prompt Skills

Learn to create prompt skills

Manifest Reference

All manifest fields

Tool Policy

Control tool access