Sensei

microsoft/sensei

Iteratively improve skill frontmatter to achieve good routing test coverage. WHEN: run sensei, sensei help, improve skill routing

2k tokens

context cost

the whole folder, loaded on every use

files

instructions only

copies elsewhere

how many repositories repackaged it

242

stars on the repo

on the repository, not the skill itself

Install

one command, takes just this skill from the repository

npx skills add https://github.com/microsoft/GitHub-Copilot-for-Azure --skill sensei

What comes with it

2 114 bytes besides the instruction

references/CONVENTIONS.md

The instruction itself

4 sections, as written by the author

Help

When user says "sensei help" or asks how to use sensei, show this:

╔══════════════════════════════════════════════════════════════════╗
║  SENSEI - Skill Frontmatter Compliance Improver                  ║
╠══════════════════════════════════════════════════════════════════╣
║                                                                  ║
║  USAGE:                                                          ║
║    Run sensei on <skill-name>              # Single skill        ║
║    Run sensei on <skill1>, <skill2>, ...   # Multiple skills     ║
║    Run sensei on all skills                # All skills          ║
║                                                                  ║
║  EXAMPLES:                                                       ║
║    Run sensei on appinsights-instrumentation                     ║
║    Run sensei on azure-ai, azure-compute                         ║
║                                                                  ║
║  WHAT IT DOES:                                                   ║
║    1. READ      - Load skill's SKILL.md and tests                ║
║    2. VERIFY    - Compare skill frontmatter with convention      ║
║    3. SCAFFOLD  - Create tests from frontmatter if missing       ║
║    4. IMPROVE   - Add WHEN: triggers                             ║
║    5. TEST      - Run tests, fix if needed                       ║
║    6. SUMMARY   - Show before/after with suggestions             ║
║    7. PROMPT    - Ask: Commit, Create Issue, or Skip?            ║
║    8. REPEAT    - Until routing tests pass                       ║
╚══════════════════════════════════════════════════════════════════╝

Main Loop

For each skill, execute this loop until the frontmatter aligns with convention, have thorough routing tests AND routing tests pass:

READ - Load plugin/skills/{skill-name}/SKILL.md, and vally eval suites in evals/{skill-name}/*.yaml.
VERIFY - Compare the skill frontmatter with conventions in CONVENTIONS. If the skill's frontmatter violates any written convention, notify the user and propose a fix to align the frontmatter with the convention.
SCAFFOLD - If evals/{skill-name}/ doesn't exist, follow instructions in vally-eval skill to scaffold a set of routing tests. The routing tests test if the skill can be invoked for target user prompts. Generate user prompts that match the target scenario of the skill's description.
IMPROVE - If the skill description doesn't already have WHEN: triggers, add them.
TEST - Follow instructions in vally-eval skill to run the routing tests. If there are failed tests, suggest fixes.
SUMMARY - Display before/after comparison with unimplemented suggestions.
PROMPT - Ask user to commit the fixes or take any alternative actions.
REPEAT - Go to step 2 (max 5 iterations per skill)

Constraints

Only modify plugin/skills/ - these are the Azure skills used by Copilot
Files in .github/skills/ should be left as is
Max 5 iterations per skill before moving on

vally-eval - vally eval suite writing guidelines

How to use it

Copy the folder

Take microsoft/sensei from the repository into ~/.claude/skills for personal use, or into .claude/skills inside a project.

Check the name does not clash

The agent identifies a skill by the name field in its header. Two skills with the same name cannot sit side by side — one of them will be ignored.