agentic-toolbox
AGENTS.md5 starsChanged 46 days agoMITView on GitHub
This repository is a shared skills home. It is not an application repo. Claude Code: you were sent here from CLAUDE.md. Stay on this file, then load what it names. Git is the source of truth. Do not write skill bodies only on a local or vendor mirror. maintainers/ holds build notes for this repository. If you are using the toolbox, skip it; routing tables never point into it. If you are working on this repository, read maintainers/AGENTS.md next; its…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 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 repository49 other files it gives its agents.
- maintainers/AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- buying-researcherSkill
- cursor-cloud-agents-whenSkill
- debug-anthropicSkill
- debug-pocockSkill
- debugSkill
- discover-the-ideaSkill
- docs-google-styleSkill
- golang-safetySkill
- golang-securitySkill
- golang-testingSkill
- grok-acpSkill
- host-config-pyinfraSkill
- lang-cppSkill
- lang-csharpSkill
- lang-cSkill
- lang-dartSkill
- lang-dockerSkill
- lang-goSkill
- lang-javaSkill
- lang-js-tsSkill
- lang-kotlinSkill
- lang-luaSkill
- lang-makefileSkill
- lang-phpSkill
- lang-powershellSkill
- lang-protobufSkill
- lang-pythonSkill
- lang-rubySkill
- lang-rustSkill
- lang-shellSkill
- lang-sqlSkill
- lang-swiftSkill
- lang-terraformSkill
- language-routerSkill
- lang-web-markupSkill
- modern-pythonSkill
- pr-reviewSkill
- sdlc-artifactsSkill
- sdlc-onboardingSkill
- security-hardeningSkill
- shell-safetySkill
- tddSkill
- tracker-sdlcSkill
- ui-craftSkill
- ux-designSkill
- verify-before-doneSkill
- yagniSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 36
Do not paste skill or SDLC bodies into this file.
- Distinctive: line 67
Before any third-party content, complete security intake: docs/INTAKE.md.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md 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
AGENTS
This repository is a shared skills home. It is not an application repo. Claude Code: you were sent here from CLAUDE.md. Stay on this file, then load what it names.
Git is the source of truth. Do not write skill bodies only on a local or vendor mirror.
maintainers/ holds build notes for this repository. If you are using
the toolbox, skip it; routing tables never point into it. If you are
working on this repository, read
maintainers/AGENTS.md next; its ## Tracker
governs.
Public repo
This is a public, open-source repo.
- Deliverables (skills, human-facing docs, code, collateral): write
for outside readers in a professional voice, per the project's style
guide (default
docs-google-style). No people's names beyond maintainer credits and git authorship, no who said what, and no private tracker workspace URLs or names, internal hostnames or IPs, credentials (see Never), or agent-session leftovers such as interview question labels. Bare ticket ids (DER-123) are fine. - Agent context (
maintainers/design/<chunk>/design records,maintainers/.agents/, othermaintainers/notes, scratch notes) can be informal: civil and credential-free, not tone-reviewed. - Commit messages, PR descriptions, and tags are public and effectively permanent: write them for outside readers.
How to load
Do not paste skill or SDLC bodies into this file. Read the named path, or pack its text into a subagent prompt per Spawn prompts.
| What | When | How |
| SDLC | Any project, ticket, chunk, or multi-agent run | Read docs/SDLC.md before implementing. |
| Roles | You need a role's name or job | Read Roles. |
| Message the operator | Any message to the operator: an ask (gate, waiver, blocker, or unclear choice), a status, or a report | Read Asking the operator and follow it. |
| Requirements | Chunk Brief, Gather: a fuzzy idea or an interview | Read skills/discover-the-idea/SKILL.md. |
| Refine | Chunk Brief, Refine: refining a brief, as the refiner agent | Read skills/yagni/SKILL.md. |
| Incoming item | Bug report, red build, unit failure, mechanical ticket | Read Entry, then Item brief. Do not load discover-the-idea. |
| Research | Purchase, product shortlist, or market research | researcher persona: read skills/buying-researcher/SKILL.md. Not an SDLC role or step: Roles. |
| UX | User stories, high-level UX, or screen mockups | designer: read skills/ux-design/SKILL.md. |
| Artifacts | Writing HLD, LLD, tickets, epics, track/roadmap, PoC, decision, changelog, PR, monthly, human how-to, AGENTS stub, or repo tracker skill (tracker-skill.md) | Read skills/sdlc-artifacts/SKILL.md and copy the matching templates/ file. A chunk brief uses discover-the-idea instead. An incoming item uses bug.md or task.md. |
| Debug | Failure while implementing a chosen fix, or unexpected behavior in Build | A new ticket that is not in Build → follow Incoming item. In Build, or after a chosen fix: read skills/debug/SKILL.md (default). Alternatives: debug-pocock, debug-anthropic. Load one. Then tdd + verify-before-done. At item Brief, the troubleshooter may load debug for root cause and must not implement. |
| Tracker | Any tracker read or write: file, claim, move, block, comment, list ready work | Read skills/tracker-sdlc/SKILL.md. Use the product repo's own ## Tracker. The skill loads the product repo's .agents/tracker/SKILL.md, or sdlc-onboarding when the setup check fails. Only the manager writes: Tracker. |
| Docs | Human how-to or agent-facing comments after Spec | Read skills/docs-google-style/SKILL.md. |
| A skill | The id applies to this turn | Read skills/<id>/SKILL.md in this repo. |
| Skill ids | Checking a skill id, its ownership, or its SHA pin | Read SOURCES.md. Agents at work never remint a skill: Roles. Changing a skill body is maintainer work: Intake plus a new security cut. |
| Language | Writing or reviewing code | Load at most one language skill per turn: iron law. Read skills/language-router/SKILL.md: its map, load-with list and no-language turns. |
| Workers | The operator picks Grok Build over ACP as a builder | Read skills/grok-acp/SKILL.md. Operator opt-in: load it only then. |
| Worker choice | The operator offloads coding or heavy multi-step work to Cursor Cloud Agent or the grok CLI | Read skills/cursor-cloud-agents-when/SKILL.md. Operator opt-in: load it only then. |
| UI craft | builder implementing or reviewing UI code | Read skills/ui-craft/SKILL.md, beside ux-design's approved mockups, not in place of ux-design. Which version, or none: the skill's ask. |
| Branches and lands | Branching an item, or landing one | Read branches-and-lands.md. |
| Subagents | Minting, resuming, or prompting a subagent or worker | Read subagents.md. |
| Knowledge | Facts about a model, vendor, API, or platform that are not a skill | Read knowledge/README.md, then knowledge/<domain>/<kind>/<slug>.md. |
| Minecraft | Building or reviewing a Bedrock structure, maze, or block plan | Not in this repo. Read skills/minecraft-builder/SKILL.md in alexderz/agentic-toolbox-misc. Block names and states are in that skill. Do not rely on remembered ids. |
Intake (security)
Before any third-party content, complete security intake: docs/INTAKE.md.
Related
- README and bundles: README.md
- SDLC (load this): docs/SDLC.md
- Knowledge: knowledge/README.md
- License: LICENSE (MIT), NOTICE
- Claude Code entry: CLAUDE.md
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.