agentleFS
syncwinsyncwin/ai-extender/skills/claude-planner/SKILL.md

claude-planner

Skill0 starsChanged 39 days agoMITView on GitHub

Plans a Claude extension before anything is built: scopes the request (intent, platform target, components, install method, naming, acronym) and designs the structure (directory tree, skill boundaries, plugin.json, router rule). Use, after the router, for any new plugin, skill, connector, agent, MCP server, or add-on; for internal tools for a team or business process; for "turn this workflow into a Claude plugin", "turn this conversation into a plugin or skill"; "what components does my extension need"; and any structural decision.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

AI Extender Planner

  1. Scope: references/scoping.md. Reuse context; ask only real gaps.
  2. Design: references/architecture.md. Include only confirmed components.
  3. Lean check: for every planned file or component ask "does the request need this?" Cut the rest; report skipped: X, add when Y. The defaults are not cut by this check: Composio stays as the default connector even when no app is named (the user may drop it), and so do version 0.0.1, the companion prompt, and behavior tests.
  4. Threat model (3 lines, in the plan): what the extension can touch (files, commands, network, accounts), what untrusted text it reads, and the worst a bad input could do. Design from ai-extender/references/standards.md §5c; a hook that guards is fail-closed, one that only nags is fail-open.
  5. One approval: show the plan in plain words and wait for a yes. This is the only approval gate before building. The plan always states: what it does; the plugin name and every skill name (one skill → named exactly like the plugin); version 0.0.1; connectors, always with Composio listed as the default (never drop it yourself, even when the extension only reads what the user gives it; say the user can remove it) and which ones the account already has; the injection guardrail when it reads outside content; how it's installed; the companion prompt's form fields, by name; anything skipped (each with when to add it, never as a question). The questions cover only what the user alone knows: who uses it, where they use Claude, and their own content; never ask whether to add a component or integration. If a reference file can't be read, plan from these rules and note it in the scratchpad only, not at the top of the reply. For a user who described the goal in everyday words with no build terms, follow the router's plain-language rule: the plan covers what it does, what they get, and how they share it, and these build details go in the scratchpad instead.
  6. Hand off to ai-extender:claude-developer. Names, acronym, metadata, licensing: ai-extender/references/standards.md.

Finish with the scratchpad (ai-extender/references/scratchpad.md).

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.