Build a reusable skill¶
Translation in Progress
This page has not yet been translated into your language. You are viewing the original English content.
Want to help? See our translation guide.
In the last lesson you hand-wrote a pipeline-level nf-test and learned which output is safe to snapshot.
Capture that discipline as a reusable skill, so CoScientist applies the same rules in every session without you re-explaining them.
1. Enhance CoScientist with your own skills.¶
CoScientist ships with built-in skills, and you can add your own. A skill is a small instruction set that CoScientist loads as session context, so it follows your guidance automatically whenever you ask for related work. Packaging the testing rules you just applied makes a good first skill.
2. Author the skill.¶
A skill lives in its own directory with a SKILL.md file: YAML frontmatter with a name and a description (both required, or the skill is skipped), followed by a markdown body of instructions.
CoScientist discovers skills from your project and user skill directories and sends them to the session as context.
A project directory like .agents/skills/ takes priority, so a repository can override a global skill.
Create the directory and the empty skill file:
Then open .agents/skills/write-nf-test/SKILL.md and add this content:
---
name: write-nf-test
description: Generate an nf-test that asserts on stable output and excludes unstable content.
---
# Write nf-test
When asked to write an nf-test for a process or a pipeline, follow the rules below.
## Steps
1. Scaffold the test under `tests/`.
For a process use a `nextflow_process` block; for a pipeline use a `nextflow_pipeline` block.
Provide representative inputs or test data so the test runs end to end.
2. Assert on deterministic output only:
- Numeric or tabular result files, such as a Salmon `quant.sf` table
- The existence of the expected output files and directories
3. Do NOT snapshot unstable content:
- Reports that embed timestamps or versions, such as MultiQC HTML
- Log files, and files that record the command line or tool version (for example Salmon `cmd_info.json` and `meta_info.json`)
- Any path or value containing the work-directory hash
- Version strings
4. Run `nf-test test` on the generated file.
When it fails, narrow the assertion rather than snapshotting unstable content, and re-run until it passes.
Keep skills small: CoScientist caps each discovered skill's context at 5 KB.
The .agents/skills/ location follows the cross-agent Agent Skills convention, so the same skill works in other compatible agents.
3. Use the skill.¶
Restart seqera ai so it discovers the new skill and loads it into the session context.
Your own skills do not appear as slash commands; the / palette is reserved for built-in, backend-provided skills.
Instead, CoScientist now follows your skill whenever you ask for a test:
It applies the stable-versus-unstable rules from your skill automatically, without you restating them.
To make the skill available across all your projects rather than one, place the same directory under ~/.agents/skills/ instead.
Checkpoint
With the skill in place, asking for a test produces one that asserts on stable output and excludes unstable content, without you re-explaining the rules.
Takeaway¶
You captured the testing discipline as a reusable skill that CoScientist loads as session context.
You and your teammates now get a consistent nf-test on demand, without re-explaining the rules each time.