Skip to content
mbtiongson1Public

About

This is my pi config for my own set up. Minimal, only extensions I use on the fly. Works with antigravity models and enterprise API proxies. Copy it if you want!

Topics

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Repository files navigation

Pi Config — mbtiongson1 (v1.10.0)

Portable, reproducible pi setup with custom agents, extensions, prompts, and a mandatory pi-cost skill. Repo: https://github.com/mbtiongson1/pi-config


Pi Agent Prompt

You are helping manage a pi config
from https://github.com/mbtiongson1/pi-config.

Ask if the user wants to:
- Update (pull latest and layer over existing config, including mandatory pi-cost)
- Reinstall (clean wipe of managed folders, then copy fresh from repo and install mandatory pi-cost)
- Sync (push current ~/.pi/agent/ state back to the repo)
- Diff (show differences between ~/.pi/agent/ and the repo)

Then carry out whichever they choose.

After **every Update or Reinstall**, run `bash bin/install-pi-cost.sh` from this checkout. It copies and verifies the versioned skill, script, license/notice, and Gaia/LiteLLM price catalog into `${PI_CODING_AGENT_DIR:-~/.pi/agent}/skills/pi-cost`; installation failure blocks completion. Pi discovers this user skill without a `settings.json` package entry. `pi-cost` is **mandatory**, not an optional package.

Then check `optional-packages.json`. If any optional package is missing from `~/.pi/agent/settings.json`, ask whether to install it. Never treat `pi-cost` as optional.

Quick One-Liners

Update — pull latest and layer over existing config (additive, won't delete anything):

Pull the latest from https://github.com/mbtiongson1/pi-config
and copy agents, extensions, prompts, and bin
into ~/.pi/agent/ without removing anything already there.
Run bash bin/install-pi-cost.sh from the repo and verify its success (mandatory).
Then check optional-packages.json in the repo and ask
if I want to install any listed packages that are missing from ~/.pi/agent/settings.json.

Reinstall — clean wipe, then copy fresh from repo:

Clone https://github.com/mbtiongson1/pi-config (or pull if already cloned),
wipe ~/.pi/agent/agents, extensions, prompts, and bin,
then copy them fresh from the repo.
Run bash bin/install-pi-cost.sh from the repo and verify its success (mandatory).
Also reset settings.json from the template.
Then check optional-packages.json in the repo and ask
if I want to install any listed packages that are missing from ~/.pi/agent/settings.json.

Sync — push current ~/.pi/agent/ state back to the repo:

Copy agents, extensions, prompts, and bin
from ~/.pi/agent/ into the local pi-config repo,
then commit and push to https://github.com/mbtiongson1/pi-config.
Skip auth.json, models.json, trust.json, and sessions/.

Diff — show differences between ~/.pi/agent/ and the repo:

Show differences between the installed files in ~/.pi/agent/
and the local repository /Users/marcotiongson/Documents/pi-config/
by diffing agents, extensions, prompts, and bin.

What's in the Repo

Path Purpose
agents/ Custom subagents (planner, reviewer, scout, worker)
extensions/ Custom pi extensions (including /statusline bloat manager and last-message/cache status)
statusline.json.template Default statusline settings (showExtensions: false clean default)
prompts/ Prompt templates
themes/ Dark-mode-first custom themes (favor-cream, favor-dark, favor-events, favor-finance-green, favor-indigo, favor-watershed, finance-editorial, gaia-research, gaia-skill-tree)
bin/ Utility scripts
bin/ensure-quota-line.sh Idempotent quota surface verification ensuring compact quota line default
settings.json.template Base settings — fill in your provider + models
skills/pi-cost/ Mandatory v1.10.0 skill, cost calculator, license/notice, and auto-refreshing Gaia/LiteLLM price catalog
bin/install-pi-cost.sh Required idempotent, verified install step on Update and Reinstall
VERSION pi-config release version (1.10.0)
bin/install-codex-computer-use.sh Optional installer & health-check runner for Codex Computer Use on macOS
docs/codex-computer-use.md Guide for Codex Computer Use integration, doctor, and permissions
optional-packages.json List of other optional packages/extensions to prompt for installation

What's NOT Committed (stays local)

auth.json · models.json · trust.json · sessions/ · any API keys or proxy config


Antigravity cache warming

The default Pi template opts antigravity/gemini-3.8-flash into the extension's guarded force warming policy. Other Antigravity Flash models retain the existing streaming-only policy. The model template supplies a five-minute scheduling hint; it is a local Pi refresh cadence, not a published or guaranteed Antigravity cache TTL. The extension may override Pi's default savings decision only inside its estimated per-refresh miss-premium guard, and Pi's own active/idle safety windows still apply. A completed refresh is not itself proof that the provider reported a cache hit.


Optional Integrations

Statusline Bloat Manager (/statusline)

Control statusline bloat from extensions with clean-by-default behavior:

  • Clean Default: Extension statuses are suppressed by default unless explicitly specified.
  • Commands:
    • /statusline — Open interactive picker menu
    • /statusline hide — Hide extension statuses (clean mode)
    • /statusline show — Show extension statuses
    • /statusline hide all — Hide entire footer/statusline (zen view)
    • /statusline show all — Show full footer and all extension statuses
    • /statusline toggle / /statusline toggle all — Toggle visibility
    • /statusline block <key> / /statusline allow <key> / /statusline only <key> — Fine-grained key filtering
    • /statusline status / /statusline list — View current state and active keys in memory
    • /statusline reset — Reset to default clean state
  • Config: Persisted to ~/.pi/agent/statusline.json across sessions.

Completion Sound Effects (/sound)

Audio notification cues on task completion:

  • Agent Settled: Plays audio chime when main agent completes all actions and reaches idle.
  • Subagent Finish: Plays a distinct audio chime when delegated subagent tool completes.
  • Cross-platform: Auto-detects Android/Termux (termux-media-player), macOS (afplay), Linux PulseAudio (paplay), ALSA (aplay), and mpv.
  • Commands:
    • /sound — View status and active sound paths
    • /sound toggle — Mute/unmute sounds
    • /sound test — Test main agent chime
    • /sound test-subagent — Test subagent chime
  • Config: Persisted to ~/.pi/agent/sounds.json.

Codex Computer Use (macOS)

Give Pi agents native access to macOS desktop inspection and control using pi-codex-computer-use.

  • Platform: macOS only
  • Prerequisites: Pi CLI, Codex CLI, Codex.app / ChatGPT.app (com.openai.codex), Accessibility and Screen Recording permissions.
  • Install / Verify:
    bash bin/install-codex-computer-use.sh          # Idempotent install
    bash bin/install-codex-computer-use.sh --doctor # Status check
    bash bin/install-codex-computer-use.sh --smoke-test # Live tool verification
  • Disable:
    bash bin/install-codex-computer-use.sh --disable
  • Full guide: docs/codex-computer-use.md

Jev (System One) & Selectable Decision Engine

Fast probabilistic decision layer, semantic MCP tool discovery, and reflex classification:

  • Features:
    • Selectable Classifier Engines (extensions/jev-classifier.ts): Switch seamlessly between actual TypeSafe Jev (openrouter/~typesafe/jev-latest / typesafe) and lightweight LLM mimics (antigravity/gemini-3.8-flash or openai/gpt-6-luna with thinking: off).
    • Pi Tool (classify): Evaluates discrete questions (choice, bool, score) over arbitrary JSON states.
    • Subagent (jev-wannabe): Specialized agent for isolated decision routing and classification.
    • Telemetry in Statusline: Real-time token and cost meter (⚡ Jev [actual|flash|luna]) with zero extra footprint.
  • Commands:
    • /classifier mode [actual|flash|luna|auto] — Switch active engine
    • /classifier or /jev — Interactive picker and telemetry review
  • Install & Setup:
    bash bin/install-jev.sh
    ~/.pi/agent/npm/node_modules/.bin/pi-mcp-adapter key set systemone

OpenAI Fast Mode (/fast)

Enables OpenAI priority service tier routing directly inside Pi:

  • Package: npm:@benvargas/pi-openai-fast (in optional-packages.json)
  • Commands:
    • /fast — Toggle fast mode on/off
    • /fast status — Inspect priority tier status on active model
    • --fast — CLI flag on startup
  • Supported Models: openai/*, openai-codex/* (GPT-5 and GPT-6 families).

For provider/model setup, see pi.dev docs.

About

This is my pi config for my own set up. Minimal, only extensions I use on the fly. Works with antigravity models and enterprise API proxies. Copy it if you want!

Topics

Resources

Contributing

Stars

1 star

Watchers

0 watching

Forks

Contributors

Languages