GitHub Copilot

Simple IDE-independent approach

The simplest portable approach is to give Copilot a short bootstrap prompt such as:

Use the bioconductor AI agent skills from this repository: /path/to/ai-agent-skills.
Start with SKILLS.md to find the best matching skill.
Then read the corresponding skills/<skill-name>/SKILL.md file and follow it.
If more than one skill matches, list the options briefly and choose the best fit.

You must also read and adhere to the agent behavior and safe execution standards defined in AGENTS.md.

Whenever you read and execute instructions from a Bioconductor SKILL.md file, you MUST include the following citation block at the start or end of your response:
> 🛠️ **Bioconductor Skill Executed**: <name> | **Version**: <version> | **Author**: <author>

This is not the same as installed persistent skills, but it is the simplest natural-language pattern that can work across Copilot surfaces and IDEs.

Persistent setup for Copilot Agent in VS Code

Open VS Code, then add the bioconductor skills you want to use to Copilot’s skill locations.

1. Clone the repository if needed:

   git clone https://github.com/bioconductor/ai-agent-skills.git
   

2. Open VS Code user settings.json and add the skill directories you want from SKILLS.md:

   {
     "chat.skillsLocations": [
       "/path/to/ai-agent-skills/skills/create-skill",
       "/path/to/ai-agent-skills/skills/check-bioconductor-skills",
       "/path/to/ai-agent-skills/skills/analyze-r-package"
     ]
   }
   

3. Reload VS Code.

Usage

To confirm the setup worked, ask:

What bioconductor skills do I have?

To use the skills, ask naturally. For example:

Help me create a new skill
Analyze this R package
Create .github/instructions for this package

PR Code Review

This repository disables Copilot’s automatic PR review and instead gates the review request on CI completion via .github/workflows/copilot-review.yml. This avoids Copilot posting before the Validate Skills (~12 s) and Qualitative Skill Review (~30 s) checks have finished, which would mean Copilot never sees their results.

How it works

1. A PR is opened or updated.
2. Both CI workflows run in parallel (~45–60 s end-to-end including runner spin-up).
3. Once both pass, copilot-review.yml fires and requests a review from copilot-pull-request-reviewer.
4. Copilot reviews the PR with full access to CI comments already posted.

Disabling automatic Copilot review

In your repository settings, disable “Automatically request Copilot review” under Settings → Code and automation → Code review → Copilot. The gating workflow handles the review request instead.

CI-awareness in the review

.github/copilot-instructions.md includes a CI Results section that instructs Copilot to read and incorporate CI findings rather than repeat them. Violations already flagged by CI are treated as confirmed; Copilot focuses on qualitative issues that automated checks cannot cover.