decorated-pi

decorated-pi is a Pi extension that adds safety gates, LSP tools, image/compaction model helpers, smarter @ file search, dynamic subdirectory AGENTS.md loading, and a few workflow quality-of-life improvements.

Install

pi install npm:decorated-pi
pi install git:github.com/lcwecker/decorated-pi
pi install /path/to/decorated-pi

Features

1. Patch Tool

Replaces Pi's built-in edit / write with a stronger patch tool:

• anchor mechanism — narrows the search range by specifying a unique string that appears before old_str, preventing mismatches in files with repeated patterns
• mtime tracking — records file modification time on read, rejects patch if the file changed since last read, preventing blind or stale edits
• explicit overwrite — offer atomic overwrite: true mode for overwrite files or full-file creation to prevent unintened overwrite

2. Smart @ File Search

Replaces Pi's default file search with a faster, project-aware search strategy:

• Uses project-aware file discovery
• Prioritizes filename-based matches for more intuitive results
• Reduces clutter from hidden, cache, and build directories
• Keeps default suggestions focused on visible project files

3. LSP Tool Suite

Based on @spences10/pi-lsp by Scott Spence (MIT License), with major additions:

• C/C++ and Lua support
• lsp_find_symbol, lsp_rename, multi-file support merged into lsp_diagnostics
• Force-sync on didChange (no stale diagnostics)

Supported languages: c/cpp, go, java, lua, python, ruby, rust, svelte, typescript

4. Auxiliary Models (Image + Compact)

Uses cheaper models for auxiliary tasks, configured via /dp-model:

• Image read fallback — when the model reads an image file, detects type via magic bytes, calls a configured vision-capable model, and replaces the read result with image analysis text (jpeg, png, gif, webp)
• Compact model — uses a configured model for context compaction (instead of the main model), auto-resumes after compaction.

5. Safety Layer

• Dangerous bash guard
  • asks for confirmation on destructive commands such as:
    • rm
    • sudo
    • svn commit/revert
    • git reset/restore/clean/push/revert
    • npm publish
    • > / 1> / 2> / &> / tee overwrite existing files
  • Hints the agent to use edit instead of write on non-empty files
• Protected paths
  • Blocks read/write access (via write/edit/read tools or cat/head/tail/grep/rg etc. bash commands) to sensitive locations such as .env, .git/, .ssh/, *.pem, *.key, etc.
• Secret redaction
  • Dual-layer detection: 40+ known-format patterns (AWS, GitHub, OpenAI, etc.) + Adjusted Shannon Entropy analysis for unknown formats. Based on opencode-secrets-protect (MIT)

6. Dynamic Subdirectory AGENTS.md / CLAUDE.md

When the agent reads or edits a file:

• discovers AGENTS.md / CLAUDE.md in the file's directory and ancestor directories
• injects newly discovered guidance into tool results

7. Extend Providers

Extend providers are registered via /login → "Use a subscription":

Provider	Base URL
Ollama Cloud	ollama.com/v1
Baidu Qianfan	qianfan.baidubce.com/v2/coding
ARK Coding	ark.cn-beijing.volces.com/api/coding/v3

Configuration

Runtime settings are stored in:

~/.pi/agent/decorated-pi.json

Module Loading

Modules can be toggled on/off. Changes take effect after /reload.

Module	Default	Effect when disabled
patch	true	Reverts to Pi's built-in edit / write tools
safety	true	No command guard, no protected path check, no secret redaction
lsp	true	All lsp_* tools unregistered — no diagnostics, hover, etc.
smart-at	true	Fallback to Pi's built-in @ file completion

Use /dp-settings to toggle, or edit the config file directly:

{
  "modules": {
    "patch": true,
    "safety": true,
    "lsp": false,
    "smart-at": true
  }
}

License

MIT