Delegate coding work from Claude Code to a local pi agent — MCP tools with mid-run steer/abort, plus per-project discipline hooks.
Call a local pi agent from inside Claude Code: Claude writes the task book and judges the result, pi writes the source and the tests.
Claude reads the code and writes the contract; pidispatch hands the writing to pi. The rows under the status line are this window's dispatches, one per row, and they disappear when the last one finishes.
Why a plugin rather than a written rule: with the rule in place, roughly 80% of the characters committed to this repo were still typed by the main model. In strict mode a PreToolUse hook denies Claude's own Write and Edit on existing product source and tells it to dispatch instead.
Dispatches go to whatever provider and model your pi is already pointed at; this plugin pins nothing.
That rail has a hole, and it is deliberate: the hook matches Write and Edit only, so the same edit made through Bash (sed -i, a heredoc) is never intercepted. It blocks editing a file yourself out of habit, not a deliberate workaround.
Run both inside Claude Code: the first registers this repository as a plugin marketplace, the second installs the plugin from it. If the install summary says Run /reload-plugins to activate., run that too. Requires Node ≥ 22 and a pi installation that is already set up; dependencies are installed automatically from the committed lockfile, so there is nothing to npm install yourself.
Runs a working copy instead of the published version. A --plugin-dir copy takes precedence over the installed one for that session, so you can test changes without uninstalling.
From the project README.
Add the radar badge to your README — it shows your project was picked up by MCP Radar and links to this page:
[](https://mcp.liqiwa.com/s/LarryStanley--pi-delegate.html)
Skill router and context picker for Claude Code, Cursor, and Aider. Auto-picks the right skill for every prompt, with prompt-injection filtering for MCP.
0xwilliamortiz/openclaude-improvedruns anywhere. uses anything
alexpilotto/uxon-aiMCP server and Claude skills for PPC landing pages, A/B experiments, and first-party conversion tracking. Build Google Ads and Meta pages, run cross-domain experiments, and pull CRO reports via the UX
bar181/bar-observatoryDeterministic, local-only audit reports for Claude Code AI agent sessions. Rust + SQLite. Zero network calls. MCP server for agents.
chris-asmussen/wardenOne MCP server that fronts many MCP servers + Skills behind a small tool set and routes to the best one, keeping an agent's context tiny. Migrate your Claude Code MCPs/skills behind it.
systempromptio/systemprompt-demoAI Governance Infrastructure — open gateway demo. Self-hosted systemprompt.io gateway in a single Rust binary: policy, audit trails, and cost attribution on every AI request, with the Systemprompt Bri
The top new MCP servers of the week, every Monday. No spam, unsubscribe anytime.