Frad's Pi Packages

English | 简体中文

Native Pi packages for reusable skills, extensions, and workflow commands.

Using packages

Skills use Pi's /skill:<name> command. Arguments after the command are appended to the skill text. Pi does not expand $ARGUMENTS or execute shell substitutions embedded in skill Markdown.

Packages that manage interactive workflows use native commands such as /keyboard, /recap, /memory, /btw, and /teammate instead of skills.

Released packages can be installed with an npm source:

pi install npm:pi-keyboard
pi install npm:@fradser/pi-memory

Every package can be used from this checkout during development:

pi install /path/to/pi-packages/packages/<name>

Packages

keyboard

Drives VIA and QMK mechanical keyboard RGB lighting to reflect Pi states, including white breathing for idle, blue breathing for thinking, green breathing for unread chat, yellow blinking for approval or questions, and red blinking for fatal errors. All updates run in memory without writing to EEPROM or Flash.

Command: /keyboard, /keyboard on, /keyboard off, /keyboard status, /keyboard test <state>

Install:

pi install npm:pi-keyboard

---

recap

Generates and displays a concise summary of current session progress above the TUI editor box. It persists across restarts and updates across parallel directory sessions.

Command: /recap, /recap on, /recap off, /recap language <lang>, /recap model <model>

Install:

pi install npm:@fradser/pi-recap

---

vision

Bridges images to text for a text-only active model through a configured vision-capable Pi model. It preserves the original session attachment and adds visual analysis only to the transient provider context.

Command: /vision, /vision model provider/model, /vision on, /vision off

Install:

pi install npm:@fradser/pi-vision

---

btw

Answers a side question in a read-only overlay without adding it to the current session history. The child Pi process may use read, grep, find, and ls to check the codebase, but cannot use bash, edit, or write.

Command: /btw <question>

Install:

pi install npm:@fradser/pi-btw

---

memory

Manages durable project memory with a /memory menu, auto-memory guidance, and background consolidation. The consolidation procedure runs in a separate child Pi process and keeps its raw work outside the active conversation.

Commands: /memory, /consolidate

Install:

pi install npm:@fradser/pi-memory

---

monitor

Runs a command in the background against an explicit result contract. It stores ordinary output outside model context and sends exactly one structured terminal result for success, failure, timeout, or a missing result.

Tools: monitor_start, monitor_stop

Skill: /skill:using-monitor

Command: /monitor

Install:

pi install npm:@fradser/pi-monitor

---

utils

Adds commands for selecting a model thinking level and recovering interrupted work. It also provides multi-session registry synchronization and directs safe git worktree add invocations into .pi/worktrees/.

Commands: /effort, /continue, /sessions

Install:

pi install npm:@fradser/pi-utils

---

agent-teams

Coordinates autonomous child Pi workers through a run-centric task graph and mailbox protocol. The team leader dispatches dependency-aware task graphs in one call, tracks progress, cancels or retries nodes, and inspects the full-screen console.

Tools: teammate_run, teammate_status, teammate_cancel, teammate_retry, teammate_message, teammate_inbox

Command: /teammate

Availability: install from this checkout. It is not currently released to npm.

---

code-context

Provides DeepWiki, Context7, and Exa retrieval tools, with clone and HTTP fetch workflows as fallbacks. The package uses direct REST calls rather than MCP sidecars.

Skills: /skill:get-context, /skill:code-context

Tools: context_deepwiki, context_context7, context_exa

Availability: install from this checkout. It is not currently released to npm.

---

mattpocock

A collection of Pi-adapted BDD, TDD, implementation, review, debugging, architecture, research, planning, handoff, teaching, and skill-writing workflows.

Skills: 27 individual skills, including /skill:bdd, /skill:tdd, /skill:implement, and /skill:code-review

Availability: install from this checkout. It is not currently released to npm.

Development

pnpm install
python3 -m pytest packages

Each package keeps its behavior scenarios in features/ and tests in tests/. Run the relevant strict TypeScript check after editing an extension:

npx tsc --noEmit --strict --skipLibCheck --target ES2022 \
  --module ESNext --moduleResolution bundler --types "" \
  packages/<name>/{src,extensions}/*.ts

Use pnpm pack --dry-run from an individual package to inspect the files that would ship.

Adding a package

1. Create packages/<name>/.
2. Add a package.json with the pi-package keyword and an explicit pi resource manifest.
3. Include every runtime resource in files and declare every imported Pi core package as a peer dependency.
4. Write the BDD scenario under features/ before implementation, then add executable tests.
5. Install the package locally with pi install /path/to/pi-packages/packages/<name> and validate its package contents.

Publishing

This repository publishes through Changesets and the GitHub Actions release workflow. Do not run recursive pnpm publish from the repository root.

For a released package change, create a Changeset, push it to main, and merge the generated version PR. The release workflow publishes only its explicit package allowlist with npm trusted publishing. A new package needs its first npm publication and trusted-publishing setup before the workflow can publish later versions.

License

Each package declares the MIT license in its own manifest. The repository does not have a separate root license file.