agent-setup
Skill77 starsChanged 20 days agoMITView on GitHub
How this repository's agent instructions are organized as Agent Skills, and how to add or change them. Use when editing AGENTS.md or anything under .agents/skills/, or when the user gives feedback about coding patterns, style, or recurring mistakes that should be captured for future sessions.
7 other projects use a version of this.
Adopted by
Most recently taken up by
- Pluviobyte/douyin-commentstheir own version, first seen Nov 15, 2026
- Hrutin-Suriya-Kotni/until-hiredtheir own version, first seen Nov 15, 2026
- Pluviobyte/saas-growth-agentstheir own version, first seen Nov 15, 2026
- NoatToan/claude-code-pluginstheir own version, first seen Nov 1, 2026
- pasky/pi-side-agentstheir own version, first seen Oct 8, 2026
- calebnewtonusc/Chewbaccatheir own version, first seen Aug 11, 2026
Other versions7 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 7
- Pluviobyte/douyin-comments★ 540same name
- Pluviobyte/saas-growth-agents★ 540same name
- pasky/pi-side-agents★ 196same name
- calebnewtonusc/Chewbacca★ 8same name
- Shad0wMazt3r/The-Scaffolding★ 3same name
- Hrutin-Suriya-Kotni/until-hired★ 2same name
- NoatToan/claude-code-plugins★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository18 other files it gives its agents.
- AGENTS.mdAGENTS.md
- config-flowSkill
- coverageSkill
- documentationSkill
- elementsSkill
- explain-planSkill
- integrationSkill
- lintSkill
- manifestSkill
- modelSkill
- performanceSkill
- pythonSkill
- release-notesSkill
- scenariosSkill
- testingSkill
- translationsSkill
- update-docsSkill
- update-versionSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
15 rules. 0 are common, 12 are distinctive, and 3 are very specific to this project.
0
Common
Most of the 4 edited versions we read keep these.
12
Distinctive
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
- Distinctive: line 23
When it gains support, delete them.
- Distinctive: line 45
Write descriptions that state the subject and the trigger, using the words someone would naturally use for the task:
- Distinctive: line 55
Avoid descriptions that only restate the skill name.
- Very specific: line 59
Create .agents/skills/<name>/SKILL.md with name and description frontmatter.
- Distinctive: line 61
Write the body as directives an agent can act on.
- Distinctive: line 66
Keep the body short and push bulk reference material into those files, so it costs nothing until needed.
- Distinctive: line 79
If a section of it has grown into a procedure, move it into a skill.
- Distinctive: line 83
When the user gives feedback about a systemic correction — a coding pattern, a style issue, an architectural decision, a recurring mistake:
- Distinctive: line 87
Check for duplicates: make sure it is not already covered elsewhere
- Distinctive: line 118
Every rule must be something an agent can act on.
- Distinctive: line 119
Remove marketing text and feature lists without guidance.
- Very specific: line 128
Keep each skill focused. If one exceeds ~500 lines, split it or move reference material into references/.
- Very specific: line 129
If one exceeds ~500 lines, split it or move reference material into references/.
- Distinctive: line 157
When a skill's claims are contradicted by the code, fix the skill in the same change.
- Distinctive: line 158
Prefer describing where things live and what the contracts are over reproducing code that will move.
“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Maintaining the agent instructions
This file governs the instruction system itself.
Layout
There is one source of truth for each kind of guidance, and everything else is a symlink:
AGENTS.md # repository-wide rules, always loaded
.agents/skills/<name>/SKILL.md # everything else, loaded on demand
CLAUDE.md -> AGENTS.md
.claude/skills -> ../.agents/skillsAGENTS.md and .agents/skills/ are the Agent Skills open standard, so Copilot, Cursor, Codex, Gemini CLI, and other compatible agents read them directly with no per-tool configuration.
Claude Code does not yet look in .agents/, which is the only reason the two symlinks exist.
When it gains support, delete them.
There are deliberately no Copilot applyTo globs, Cursor rules, or reusable prompt files.
Those were per-tool formats for the same content; skills replaced all of them.
Permission grants stay out of the repository
.claude/settings.json and .claude/settings.local.json are gitignored, and no equivalent for any other agent belongs in version control either.
An allowlist entry for anything that executes project code — pytest, uv run, the gate runner — is arbitrary code execution for an agent that can also edit that code.
Committing one silently grants that to everyone who clones the repository, in environments that may not be sandboxed like the author's.
Whether to pre-approve a command is a per-developer, per-environment trust decision, so leave it to the developer.
Skills may describe what a command does and when to run it. They must not try to arrange for it to run without a prompt.
How skills load
Agents read only each skill's name and description at startup, then load the full body when a task matches.
The description decides whether the skill is used at all.
A perfect skill with a vague description never fires.
Write descriptions that state the subject and the trigger, using the words someone would naturally use for the task:
description: User-facing strings in translations/en.json — sensor and device
display names, entity translation keys, config flow errors, exception and
repair issue messages. Use when adding or renaming any user-visible string,
entity name, or output sensor.Name concrete files, directories, and domain nouns, but never a list of element kinds (see Don't enumerate groups). Avoid descriptions that only restate the skill name.
Adding a skill
- Create
.agents/skills/<name>/SKILL.mdwithnameanddescriptionfrontmatter. Thenamemust match the directory. - Write the body as directives an agent can act on.
- Nothing else. Discovery is automatic in every agent, and the Claude Code symlink already covers the whole tree.
A skill directory may also hold scripts/, references/, and assets/ for material the body links to.
Keep the body short and push bulk reference material into those files, so it costs nothing until needed.
Executable tooling that humans also run belongs in tools/, not inside a skill.
Choosing where guidance goes
| Put it in | When |
AGENTS.md | It applies to essentially every change, or an agent must know it up front |
| A skill | It applies to one area, task, or workflow |
docs/ | A human needs the explanation, not just the directive |
AGENTS.md is loaded in full on every request, so keep it short.
If a section of it has grown into a procedure, move it into a skill.
Self-maintenance
When the user gives feedback about a systemic correction — a coding pattern, a style issue, an architectural decision, a recurring mistake:
- Identify scope: which area does it belong to?
- Find the target: the matching skill, or
AGENTS.mdif it is universal - Check for duplicates: make sure it is not already covered elsewhere
- Write it as a directive: something an agent can act on, not a description of the problem
Content guidelines
Use semantic line breaks
One sentence per line, with optional breaks at clause boundaries for clarity.
Don't enumerate groups
Describe the category pattern rather than listing members. Enumeration creates brittle rules that go stale as the codebase changes.
<!-- ❌ Bad: Enumeration -->
Each element (Battery, Grid, Load, Solar, Node) must have...
<!-- ✅ Good: Pattern description -->
Each element type registered in ElementType must have...The test for good grouping: if you cannot identify the group without enumerating it, it is not a well-defined group.
Descriptions may name concrete files, directories, and tools, because that helps an agent decide whether the skill is relevant.
Element kinds are not an exception: describe them as "every type registered in ElementType" in a description as well as a body, so adding an element type never leaves a skill silently stale.
Actionable content
Every rule must be something an agent can act on. Remove marketing text and feature lists without guidance.
Explanatory background
Background context is allowed when it improves decision-making. "We use uv" is useful context; "uv is fast" is marketing.
Concise
Keep each skill focused.
If one exceeds ~500 lines, split it or move reference material into references/.
DRY
Link to documentation for detailed explanations. Skills contain directives; docs contain explanations.
What makes a good rule
| ✅ Good rule | ❌ Bad rule |
"Use str | None not Optional[str]" | "Python has several ways to express optional types" |
| "Keep try blocks minimal" | "Error handling is important" |
"Use asyncio.gather() for multiple awaits" | "Async programming has many benefits" |
| "Elements are registered in ELEMENT_TYPES" | "Battery, Grid, Load, PV, Node are elements" |
When to update skills vs documentation
| Update a skill when... | Update docs when... |
| Adding a new directive | Explaining why a pattern exists |
| Changing a coding standard | Providing extended examples |
| Adding anti-patterns to avoid | Documenting architecture decisions |
| Agent-specific behavioral guidance | Writing human-readable tutorials |
Keeping skills honest
Skills that describe code drift as the code changes, and a confidently wrong skill is worse than no skill. When a skill's claims are contradicted by the code, fix the skill in the same change. Prefer describing where things live and what the contracts are over reproducing code that will move.
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.

