skills
Copilot instructions1 starsChanged 8 months agoMITView on GitHub
Personal skills repository overview and agent operating guidelines
Where files are headingNew Copilot instructions files have more rules about branches and pull requests, and fewer about code style.
Of the Copilot instructions files we read, the 2,397 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- global/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- browser-toolsSkill
- clone-with-hashSkill
- decompose-taskSkill
- diff-checkSkill
- ios-safari-debugSkill
- market-researchSkill
- publish-artifactSkill
- spec-templateSkill
The file
We split this copilot-instructions.md into the rules it gives an agent
9 rules. About 0 are common, 6 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈6
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 18
When asked to set up, repair, synchronize, or explain agent customizations, read BOOTSTRAP.md.
- Very specific: line 24
Use @lead as the main orchestrator by default.
- Distinctive: line 34
Before editing anything, identify the intended local checkout or worktree from the user's path, the current workspace, or the repository's …
- Very specific: line 38
If the user names a checkout or worktree explicitly, use it.
- Distinctive: line 50
Use the single sync script:
- Distinctive: line 74
If there are compilation/typecheck errors, fix those first.
- Distinctive: line 75
Prefer targeted tests over full suites.
- Distinctive: line 81
Keep changes small and reviewable.
- Distinctive: line 82
Avoid personal-project-specific assumptions (hard-coded dev commands, project-specific toolchains).
≈ “Common” is an estimate: at least 1 in 5 of the 3,854 Copilot instructions files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Skills Repository (Entry Point)
This repository is the source of truth for reusable:
- Agents (
.github/agents/*.agent.md) - Prompts (
.github/prompts/*.prompt.md) - Skills (
.github/skills/*/SKILL.md) - Instructions (this file)
- MCP definitions (
mcp/servers.json)
It is an entry point: start here, then move to the appropriate target repository to implement a task.
Machine Bootstrap
When asked to set up, repair, synchronize, or explain agent customizations, read BOOTSTRAP.md. Preview with ./setup.sh bootstrap --dry-run, apply only when requested, and finish with ./setup.sh validate. Never copy credentials into Git; MCP secrets stay in environment variables or client OAuth stores.
Edit canonical files under .github/ and mcp/servers.json. Do not edit .generated/ or projected user-directory links.
Default Entry Point Agent
Use @lead as the main orchestrator by default.
@lead is responsible for:
- Turning a request into an ordered task breakdown (in chat or in a tracking issue)
- Delegating implementation to
@engineer - Requesting review from
@reviewer - Pulling in
@plannerfor decomposition and sequencing when helpful
Where to Work
Before editing anything, identify the intended local checkout or worktree from the user's path, the current workspace, or the repository's Git remote. Do not depend on a centrally maintained repo-to-path mapping.
If the user names a checkout or worktree explicitly, use it. Otherwise, inspect
the current workspace first, then look for a nearby checkout whose origin
remote matches the requested repository. Ask before cloning a new checkout or
switching a dirty working tree.
Legacy VS Code round-trip sync
VS Code Insiders stores user agents/prompts/instructions as flat files under:
~/.config/Code - Insiders/User/prompts/
This repo stores canonical versions under .github/.
Use the single sync script:
- Repo → user-data preview:
./sync.sh to-userdata --dry-run - Repo → user-data apply:
./sync.sh to-userdata --apply - User-data → repo preview:
./sync.sh from-userdata --dry-run - User-data → repo apply:
./sync.sh from-userdata --apply
For normal cross-harness installation, use ./setup.sh bootstrap instead.
To copy these definitions into another working repo’s .github/:
./sync.sh to-repo /absolute/path/to/repo
Operating Guidelines (VS Code–inspired)
These guidelines are adapted from the VS Code repo’s Copilot instructions (as a style reference), but are generalized so they apply across the repos you work in.
Default approach
- Find the right repo + folder (use the mapping table).
- Search before coding (semantic search for concepts, grep for exact strings, follow imports).
- Prefer existing patterns over inventing new ones.
- Validate incrementally (compile/lint first, then tests).
Validation discipline
- If there are compilation/typecheck errors, fix those first.
- Prefer targeted tests over full suites.
- Don’t declare work “done” while the build/watch pipeline is red.
Quality and safety
- Don’t commit or push unless explicitly requested.
- Keep changes small and reviewable.
- Avoid personal-project-specific assumptions (hard-coded dev commands, project-specific toolchains).
- For user-facing changes: always request explicit manual verification steps when automated verification is not available.
Typical daily workflows this repo should support
- Issue triage (summarize, repro ask, labels, dedupe, route to owner)
- PR authoring (clear description, tests, screenshots if applicable)
- PR review (correctness, performance, layering/architecture, test coverage)
- Release notes (summarize commit/PR history into user-facing bullets)
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

