claude-developer
Skill0 starsChanged 39 days agoMITView on GitHub
Builds the components of a Claude extension: skills, connectors (MCP config), agents, hooks, commands, MCP servers, LSP servers, output styles, and add-ons. Use for "make me a skill that formats commit messages", "build the planned skill", "add a connector for Notion to this plugin", "make an agent", "write a hook that blocks edits to .env", "build an MCP server", "hook this up to X", "extend this plugin", or "test this skill". Drafts SKILL.md, .mcp.json, agent files, and evals.
- Reads credentials
In this repository9 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/claude-developer/references/agents.mdAGENTS.md
- .claude/CLAUDE.mdCLAUDE.md
- ai-extenderSkill
- claude-maintainerSkill
- claude-packagerSkill
- claude-plannerSkill
- claude-reviewerSkill
- prompt-builderSkill
The file
We split this skill into the rules it gives an agent
6 rules. About 0 are common, 1 is distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 44
Verify field names and limits against ai-extender/references/claude-platform-facts.md (refresh via claude-docs-sync.md when stale or …
- Very specific: line 46
Secure by default: every component follows ai-extender/references/standards.md §5c (exec-form commands, quoted ${CLAUDE_PLUGIN_ROOT}, path …
- Distinctive: line 48
Every skill, agent, and connector you write ends with a scratchpad step.
- Very specific: line 49
Every extension gets its companion prompt: write the fields agreed in the plan as a spec and run companion_prompt.py <dir> --spec …
- Very specific: line 50
Run the de-duplication pass (claude-maintainer/references/versioning.md Part 1) on each skill written.
- Very specific: line 51
Write the behavior tests with the skills, not after: at least 3 cases per skill a user can call, in evals/ (references/testing.md).
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
AI Extender Developer
Build only what the plan confirmed. No approved plan in context: hand to ai-extender:claude-planner first. Load the one reference for the component:
| Component | Reference |
| Skill | references/skills.md |
Connector / .mcp.json | references/connectors.md |
| Custom MCP server | references/mcp-servers.md |
| Agent | references/agents.md |
| Hook | references/hooks.md |
| Command (legacy) | references/commands.md |
| Add-on for another extension | references/addons.md |
LSP, output styles, monitors, themes, workflows, bin/, plugin settings, channels | references/components-extra.md |
| Mods (JavaScript hooks module) | references/mods.md |
Local MCP bundle (.mcpb) | references/mcpb-bundles.md |
Skills outside plugins (.claude/skills, claude.ai, API) and their limits | references/upload-and-api.md |
| Testing and description tuning | references/testing.md (scripts/aggregate_results.py; agents ai-extender:claude-grader, ai-extender:claude-comparator) |
| Companion prompt (Prompt Builder JSON) | ai-extender/references/companion-prompt.md; ../claude-packager/scripts/companion_prompt.py --help |
| New project skeleton | scripts/scaffold_extension.py --help (also writes a default companion prompt) |
| MCP server skeleton (Node/Python) | scripts/scaffold_mcp_server.py --help |
| Eval review page | scripts/eval_report.py --help |
| Injection guardrail, small local models, Docker | references/local-models.md; scaffolder adds the guardrail by default with connectors (--no-guardrail) |
Rules
- Lean first: least code and text that meets the request; reuse existing tools, connectors, skills. Report
skipped: X, add when Y. - Verify field names and limits against
ai-extender/references/claude-platform-facts.md(refresh viaclaude-docs-sync.mdwhen stale or doubtful). - Names, acronym, metadata:
ai-extender/references/standards.md. - Secure by default: every component follows
ai-extender/references/standards.md§5c (exec-form commands, quoted${CLAUDE_PLUGIN_ROOT}, path containment, fail-closed guards, no secrets); the reference templates already do. - Copy the shared foundation (scratchpad, efficiency ladder, lean build, security; standards §5) into generated extensions when relevant.
- Every skill, agent, and connector you write ends with a scratchpad step.
- Every extension gets its companion prompt: write the fields agreed in the plan as a spec and run
companion_prompt.py <dir> --spec <spec.json>(the scaffold's default is a starting point, not the finish). - Run the de-duplication pass (
claude-maintainer/references/versioning.mdPart 1) on each skill written. - Write the behavior tests with the skills, not after: at least 3 cases per skill a user can call, in
evals/(references/testing.md). - Enforcement and quality bar:
ai-extender/references/standards.md§5b and §5d.
Next: ai-extender:claude-reviewer. Finish with the scratchpad (ai-extender/references/scratchpad.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.