Skip to content
Agentmart
Submit a tool

pi-coding-agent

Guides agents using Pi coding-agent: install, authenticate, run, configure, automate, and extend the pi CLI.

Skill by minfengyuan

IndexedMIT★ 1
Write a review

Install

Works with Claude Code, Claude apps, Codex, Cursor, Gemini CLI, GitHub Copilot, VS Code, OpenCode, Goose, Windsurf. Read the repository before you give any tool access to your data.

npx agentmarket install skill/minfengyuan-pi-coding-agent

Asks which of your tools to add it to (Claude Code, Codex, Cursor, VS Code, Windsurf, Gemini CLI, Claude Desktop) and shows the change first. See https://agentmart.dev/cli.

Get the weekly diff

The weekly diff: one email a week with new agent tools, notable updates and grade changes, plus security alerts for tools you follow. Double opt-in, one-click unsubscribe.

We use your email only for this newsletter. Privacy

About

Guides agents using Pi coding-agent: install, authenticate, run, configure, automate, and extend the pi CLI. Use for interactive, print, JSON, RPC, or SDK workflows; providers, models, settings, resources, sessions, compaction, extensions, packages, and Pi security.

Use cases: security, agent-building

No reviews yet. Installed it? Reviews come from verified installers only. How reviews work

Permissions

What this skill can do and what it asks for, from its SKILL.md. Checked Oct 25, 2026.

Tools
No tools found
Secrets
None asked for
Install scripts
None found

Network hosts it talks to 1

  • codeload.github.com · found in install command

Files it touches

  • ~/.claude/skills/pi-coding-agent

What's inside

Skill name
pi-coding-agent
Description
Guides agents using Pi coding-agent: install, authenticate, run, configure, automate, and extend the pi CLI. Use for interactive, print, JSON, RPC, or SDK workflows; providers, models, settings, resources, sessions, compaction, extensions, packages, and Pi security.
License
MIT
Compatibility
Requires the Pi coding-agent CLI or the @earendil-works/pi-coding-agent package. Follow the pinned reference snapshot for version-specific behavior.
Metadata
source: https://github.com/earendil-works/pi/tree/main/packages/coding-agent/docs/ · source_commit: dcfe36c79702ec240b146c45f167ab75ecddd205 · reference_index: references/INDEX.md
Instructions
56 lines in SKILL.md · view on GitHub

Files (46)

  • SKILL.md
  • metadata.json
  • INDEX.md · references/
  • configuration.md · references/core/
  • resources.md · references/core/
  • sessions.md · references/core/
  • usage.md · references/core/
  • development.md · references/development/
  • extensions.md · references/development/
  • providers.md · references/development/
  • tui.md · references/development/
  • json.md · references/integration/
  • rpc.md · references/integration/
  • sdk.md · references/integration/
  • README.md · references/platform/
  • containerization.md · references/safety/
  • security.md · references/safety/
  • compaction.md · references/upstream/
  • containerization.md · references/upstream/
  • custom-provider.md · references/upstream/
  • development.md · references/upstream/
  • docs.json · references/upstream/
  • extensions.md · references/upstream/
  • index.md · references/upstream/
  • json.md · references/upstream/
  • keybindings.md · references/upstream/
  • models.md · references/upstream/
  • packages.md · references/upstream/
  • prompt-templates.md · references/upstream/
  • providers.md · references/upstream/
  • quickstart.md · references/upstream/
  • rpc.md · references/upstream/
  • sdk.md · references/upstream/
  • security.md · references/upstream/
  • session-format.md · references/upstream/
  • sessions.md · references/upstream/
  • settings.md · references/upstream/
  • shell-aliases.md · references/upstream/
  • skills.md · references/upstream/
  • terminal-setup.md · references/upstream/
  • termux.md · references/upstream/
  • themes.md · references/upstream/
  • tmux.md · references/upstream/
  • tui.md · references/upstream/
  • usage.md · references/upstream/
  • windows.md · references/upstream/

SKILL.md (first lines)

# Pi coding-agent

## Purpose

Use this skill as a routing layer for Pi coding-agent tasks. Do not load every reference at once. Start with the smallest matching guide in `references/INDEX.md`, then read an upstream snapshot only when an exact option, event, type, or protocol detail is required.

## Operating workflow

1. **Classify the request**: daily CLI use, configuration, resources, sessions, automation, extension development, UI development, security, or platform setup.
2. **Read one focused reference** from `references/` before making version-sensitive claims.
3. **Inspect only the environment needed for the action**: check `pwd`, `pi --version`, `pi --help`, project `AGENTS.md`, or related files when they can change the result. Do not run an environment checklist for explanation-only or narrowly scoped tasks.
4. **Choose the smallest interface**: use the CLI for normal work, JSON for an event stream, RPC for a process/client boundary, and the SDK for in-process Node.js integration.
5. **Act with explicit scope**: make cwd, provider/model, session behavior, and project trust explicit only when they affect correctness or safety.
6. **Verify the behavior changed or configured by the task** and report failures without exposing credentials. Broaden verification only when the change or new evidence justifies it.

## Fast task router

| User intent | Read first |
|---|---|
| Install, authenticate, start Pi, use files, shell commands, slash commands, or CLI flags | `references/core/usage.md` |
| Configure API keys, providers, models, settings, thinking, retries, or keybindings | `references/core/configuration.md` |
| Create or use Skills, prompt templates, or Pi packages | `references/core/resources.md` |
| Resume, fork, clone, navigate, inspect, or compact sessions | `references/core/sessions.md` |
| Consume structured events from a one-shot run | `references/integration/json.md` |
| Build a language-agnostic or subprocess client | `references/integration/rpc.md` |
| Embed Pi in a Node.js application | `references/integration/sdk.md` |
| Add tools, commands, hooks, or lifecycle behavior | `references/development/extensions.md` |
| Add a model/provider, proxy, OAuth, or custom stream | `references/development/providers.md` |
| Build custom TUI, themes, or keybindings | `references/development/tui.md` |
| Run untrusted or unattended work | `references/safety/security.md`, then `containerization.md` |
| Fix terminal, tmux, Windows, Termux, or shell behavior | `references/platform/README.md` |

## Essential facts

- The normal built-in tools are `read`, `bash`, `edit`, and `write`; `grep`, `find`, and `ls` are also available through tool options.
- Interactive mode is `pi`; one-shot mode is `pi -p`; JSON events use `pi --mode json`; JSONL client integration uses `pi --mode rpc`.
- `@path` includes a file in the prompt. In interactive mode, `!command` runs bash and sends its output to the model; `!!command` runs it without adding the output to model context.
- Context files include global `~/.pi/agent/AGENTS.md` and project/ancestor `AGENTS.md` or `CLAUDE.md`. `/reload` reloads resources and context files.
- Sessions persist under `~/.pi/agent/sessions/` by default. Use `/resume`, `/new`, `/tree`, `/fork`, `/clone`, and `/compact`, or their CLI equivalents.
- Project Trust controls loading of project-local Pi resources; it is **not** a sandbox. Pi runs with the permissions of the launching user and has no built-in sandbox.
- Skills, extensions, packages, model configuration, and project-local resources can execute or instruct arbitrary actions. Review them before enabling them.

## Safety rules

- Never print, paste, or commit API keys, OAuth tokens, `auth.json`, or secret environment values.
- Do not treat a trusted project, a skill, or a package as safe merely because Pi loaded it.
- For untrusted repositories, generated code that will run unattended, or destructive automation, recommend a container, VM, micro-VM, or policy-controlled sandbox and mount only required paths or credentials.
- Do not invent a Pi API or configuration field. Read the focused reference or the matching file in `references/upstream/`.
- Do not use RPC built-in TUI commands as if they were protocol commands; RPC clients must consume JSONL responses/events and handle `streamingBehavior` while a run is active.
- Do not load `references/upstream/` wholesale. It is a version-pinned fallback reference, not default context.

## Completion

Before claiming completion, verify the parts of Pi behavior the request actually changed or depends on. For extension or SDK work, run the relevant TypeScript/tests and read the exact upstream reference for the API being used. Do not require unrelated provider, session, trust, or interface checks for a task that does not touch them.

From minfengyuan/harness-knowledge-base, MIT.

More from minfengyuan/harness-knowledge-base

Signals

Stars
1
Forks
0
Open issues
0
Last commit (this folder)
Sep 23, 2026 (1 month ago)

From GitHub (fetched Oct 24, 2026).

Details

License
MIT
First seen
Oct 24, 2026
Last checked
Oct 24, 2026

More from minfengyuan

Cite this page

pi-coding-agent skill. Agent Mart. https://agentmart.dev/skills/minfengyuan-pi-coding-agent (last updated 2026-10-24)