@code-yeongyu/senpi 2026.5.15 → 2026.5.16

package/docs/agents.md

@@ -1,348 +0,0 @@
-> pi can create custom agents. Ask it to build one for your use case.
-
-# Agents
-
-Agents are typed subagent profiles that control which tools a spawned task can use and what system prompt it receives. Use agents to enforce read-only exploration, restrict dangerous operations, or create specialized task runners.
-
-## Table of Contents
-
-- [Built-in Agents](#built-in-agents)
-- [Using Agents](#using-agents)
-- [Custom Agents](#custom-agents)
-  - [Locations](#locations)
-  - [Frontmatter](#frontmatter)
-  - [Example](#example)
-- [Permissions](#permissions)
-  - [Permission Actions](#permission-actions)
-  - [Permission Config](#permission-config)
-  - [Evaluation Order](#evaluation-order)
-  - [Ask Mode](#ask-mode)
-- [Configuration](#configuration)
-  - [Directory Setup](#directory-setup)
-  - [Writing an Agent File](#writing-an-agent-file)
-  - [Settings](#settings)
-  - [Priority Order](#priority-order)
-  - [Quick Start Recipes](#quick-start-recipes)
-- [How It Works](#how-it-works)
-- [Known Limitations](#known-limitations)
-
-## Built-in Agents
-
-| Agent | Tools | System Prompt | Description |
-|-------|-------|---------------|-------------|
-| `general` | All except `task`, `todowrite` | Default pi prompt | General-purpose agent for parallel work |
-| `explore` | `read`, `grep`, `find`, `ls`, `bash` | File search specialist | Read-only codebase exploration |
-
-### general
-
-The default agent for complex multi-step tasks. It can use all tools except `task` and `todowrite` (to prevent nested subagent spawning and todo list conflicts).
-
-Permission config:
-
-```json
-{
-   "*": "allow",
-   "task": "deny",
-   "todowrite": "deny"
-}
-```
-
-### explore
-
-A read-only agent optimized for fast codebase exploration. It can search, read, and list files but cannot modify anything.
-
-Permission config:
-
-```json
-{
-   "read": "allow",
-   "grep": "allow",
-   "find": "allow",
-   "ls": "allow",
-   "bash": "allow"
-}
-```
-
-System prompt appended:
-
-```
-You are a file search specialist. You excel at thoroughly navigating and exploring codebases.
-Guidelines:
-- Search file contents by regex or literal pattern when you need to locate usages or definitions
-- Read files directly when you already know the path
-- List directory contents to build a map of unfamiliar areas
-- Return file paths as absolute paths
-- Do not create any files or modify the system state
-Complete the search request efficiently and report findings clearly.
-```
-
-## Using Agents
-
-Pass `agent_type` when spawning a task:
-
-```typescript
-task(agent_type="explore", prompt="Find all files importing React", run_in_background=true)
-task(agent_type="general", prompt="Refactor the auth module", run_in_background=false)
-```
-
-Without `agent_type`, the task tool works exactly as before (no restrictions, full backward compat).
-
-## Custom Agents
-
-### Locations
-
-Pi loads custom agents from Markdown files in:
-
-- Global: `~/.senpi/agent/**/*.md` and `~/.senpi/agents/**/*.md`
-- Project: `.senpi/agent/**/*.md` and `.senpi/agents/**/*.md` (in current working directory)
-
-Both `agent/` and `agents/` directories are scanned recursively. The agent name is derived from the filename: `my-agent.md` becomes `my-agent`.
-
-**Name collision resolution:** Project-local agents override global agents with the same name.
-
-### Frontmatter
-
-Custom agents use YAML frontmatter (same format as skills):
-
-| Field | Type | Required | Description |
-|-------|------|----------|-------------|
-| `description` | string | No | What this agent does and when to use it |
-| `mode` | string | No | `"subagent"`, `"primary"`, or `"all"` (default: `"all"`) |
-| `model` | string | No | Model ID for this agent (e.g., `"anthropic/claude-haiku-4-5"`) |
-| `temperature` | number | No | Sampling temperature 0-2 |
-| `tools` | object | No | Permission config (see below) |
-| `disable` | boolean | No | When `true`, agent is not loaded |
-
-**Mode values:**
-- `"subagent"` - Only runs as a task subagent (spawned via `task()`)
-- `"primary"` - Only runs in the main session (not as a subagent)
-- `"all"` - Can run in both contexts
-
-### Example
-
-Create `.senpi/agents/readonly.md`:
-
-```markdown
----
-description: Read-only agent for safe codebase exploration
-mode: subagent
-tools:
-   read: allow
-   grep: allow
-   find: allow
-   ls: allow
-   bash: deny
-   write: deny
-   edit: deny
----
-
-You are a read-only exploration agent. You can search and read files but cannot modify anything.
-
-Guidelines:
-- Use grep to search for code patterns
-- Use read to examine file contents
-- Use find and ls to navigate the filesystem
-- Never suggest file modifications
-- Report findings clearly and concisely
-```
-
-Use it:
-
-```typescript
-task(agent_type="readonly", prompt="Find all TODO comments in the codebase", run_in_background=true)
-```
-
-## Permissions
-
-### Permission Actions
-
-Three actions control tool access:
-
-- `allow` - Tool executes normally
-- `deny` - Tool blocked, LLM receives error message explaining the restriction
-- `ask` - In interactive mode, user is prompted to allow/deny. In non-interactive mode (json/print), auto-denied.
-
-### Permission Config
-
-The `tools` frontmatter field accepts two formats.
-
-**Simple format** - single action for all patterns:
-
-```yaml
-tools:
-   read: allow
-   bash: deny
-   edit: ask
-```
-
-**Nested format** - pattern-specific rules (for future extensibility):
-
-```yaml
-tools:
-   read:
-      "*": allow
-      "*.env": ask
-   bash: deny
-```
-
-Currently, pattern matching is not implemented for specific paths. The `"*"` pattern applies to all uses of that tool.
-
-### Evaluation Order
-
-Permissions are evaluated using `findLast` semantics: **later rules override earlier ones**.
-
-Merge order (later wins):
-
-1. Global defaults from `settings.json` (`agentDefaults.permission`)
-2. Built-in agent permissions (for built-in agents)
-3. Custom agent permissions (from frontmatter `tools`)
-
-Example: If global defaults set `"bash": "ask"` but a custom agent sets `"bash": "allow"`, the agent's explicit permission wins.
-
-### Ask Mode
-
-In **interactive mode**, when a tool with `"ask"` permission is called:
-
-1. TUI displays a select prompt: "Allow once / Allow always / Deny"
-2. "Allow once" permits this single call
-3. "Allow always" persists a matching approval rule to `.senpi/permissions-approved.jsonl`, so matching calls are auto-allowed after reload too
-4. "Deny" blocks with an error message
-
-In **non-interactive mode** (json or print output), `ask` permissions are auto-denied with an explanatory message.
-
-## Configuration
-
-### Directory Setup
-
-Create your agent directory structure:
-
-```
-# Project-local agents (recommended)
-.senpi/
-   agents/
-      my-agent.md
-      code-reviewer.md
-
-# Global agents (shared across projects)
-~/.senpi/
-   agents/
-      my-global-agent.md
-```
-
-Both `agent/` and `agents/` subdirectories are scanned recursively.
-
-### Writing an Agent File
-
-Agent files are Markdown with YAML frontmatter. The body (after `---`) becomes the agent's system prompt.
-
-```markdown
----
-description: Strict code reviewer that only reads and analyzes
-mode: subagent
-model: anthropic/claude-sonnet-4-20250514
-temperature: 0.3
-tools:
-   read: allow
-   grep: allow
-   find: allow
-   ls: allow
-   bash: allow
-   edit: deny
-   write: deny
-   task: deny
----
-
-You are a code review specialist. Analyze code for bugs, security issues, and style violations.
-
-Rules:
-- Never suggest modifying files directly
-- Always reference specific line numbers
-- Categorize issues as: critical, warning, or suggestion
-```
-
-### Settings
-
-Configure default permissions and model for all agents in `settings.json`.
-
-| Location | Scope |
-|----------|-------|
-| `~/.senpi/agent/settings.json` | Global (all projects) |
-| `.senpi/settings.json` | Project (overrides global) |
-
-Add the `agentDefaults` key:
-
-```json
-{
-   "agentDefaults": {
-      "permission": {
-         "write": "ask",
-         "edit": "ask",
-         "bash": "allow",
-         "read": "allow"
-      },
-      "model": "anthropic/claude-haiku-4-5"
-   }
-}
-```
-
-| Setting | Type | Default | Description |
-|---------|------|---------|-------------|
-| `agentDefaults.permission` | object | `{}` | Default tool permissions for all agents (see [Permission Actions](#permission-actions)) |
-| `agentDefaults.model` | string | - | Default model ID for spawned agents |
-
-These defaults are the **lowest priority** layer. Agent-specific permissions (from frontmatter `tools`) override them.
-
-### Priority Order
-
-Permissions are resolved in this order (later wins):
-
-1. `settings.json` -> `agentDefaults.permission` (lowest priority)
-2. Built-in agent permissions (`general`, `explore` configs)
-3. Custom agent frontmatter `tools` (highest priority)
-
-Example: Global settings set `bash: ask`, but your custom agent sets `bash: allow` -> bash is allowed for that agent.
-
-### Quick Start Recipes
-
-**Read-only explorer:**
-```typescript
-task(agent_type="explore", prompt="Find all error handling patterns", run_in_background=true)
-```
-
-**Custom strict reviewer:**
-1. Create `.senpi/agents/reviewer.md` (see [Writing an Agent File](#writing-an-agent-file))
-2. Use: `task(agent_type="reviewer", prompt="Review changes in src/auth/", run_in_background=false)`
-
-**Restrict all agents by default:**
-Add to `~/.senpi/agent/settings.json`:
-```json
-{
-   "agentDefaults": {
-      "permission": {
-         "write": "ask",
-         "edit": "ask"
-      }
-   }
-}
-```
-Now every agent must get user confirmation before writing or editing files.
-
-## How It Works
-
-The agent system is implemented as a builtin extension that intercepts the task tool:
-
-1. **Environment variable** - When `task(agent_type="...")` is called, the agent type is passed via `SENPI_AGENT_TYPE` environment variable to the subprocess
-2. **Registry lookup** - On session start, the extension scans `~/.senpi/` and `.senpi/` for agent definitions, merges with built-in agents, and resolves by name
-3. **Tool filtering** - `setActiveTools()` removes denied tools from the LLM's tool list entirely (they don't appear in the API call)
-4. **Defense in depth** - A `tool_call` event handler acts as backup, catching any tools that slip through (e.g., added dynamically after session start)
-5. **System prompt** - The `before_agent_start` event appends the agent's custom prompt to the existing system prompt (doesn't replace it)
-6. **Permission evaluation** - Each tool call is checked against the merged permission rules (global defaults + agent config)
-
-## Known Limitations
-
-- `disable: true` in custom agents does not currently remove built-in agents (v1 limitation)
-- Bash permission bypass: denying `edit` but allowing `bash` still permits `echo > file` via bash redirection
-- No sub-pattern matching for bash commands (entire bash tool allow/deny only, no per-command filtering)
-- `ask` mode in non-interactive (json/print) mode auto-denies without user input
-- No dynamic permission changes at runtime (permissions are immutable after agent start)
-- No agent permission inheritance chains (flat configs only, no "extends" mechanism)

package/examples/extensions/subagent/README.md

@@ -1,172 +0,0 @@
-# Subagent Example
-
-Delegate tasks to specialized subagents with isolated context windows.
-
-## Features
-
-- **Isolated context**: Each subagent runs in a separate `pi` process
-- **Streaming output**: See tool calls and progress as they happen
-- **Parallel streaming**: All parallel tasks stream updates simultaneously
-- **Markdown rendering**: Final output rendered with proper formatting (expanded view)
-- **Usage tracking**: Shows turns, tokens, cost, and context usage per agent
-- **Abort support**: Ctrl+C propagates to kill subagent processes
-
-## Structure
-
-```
-subagent/
-├── README.md            # This file
-├── index.ts             # The extension (entry point)
-├── agents.ts            # Agent discovery logic
-├── agents/              # Sample agent definitions
-│   ├── scout.md         # Fast recon, returns compressed context
-│   ├── planner.md       # Creates implementation plans
-│   ├── reviewer.md      # Code review
-│   └── worker.md        # General-purpose (full capabilities)
-└── prompts/             # Workflow presets (prompt templates)
-    ├── implement.md     # scout -> planner -> worker
-    ├── scout-and-plan.md    # scout -> planner (no implementation)
-    └── implement-and-review.md  # worker -> reviewer -> worker
-```
-
-## Installation
-
-From the repository root, symlink the files:
-
-```bash
-# Symlink the extension (must be in a subdirectory with index.ts)
-mkdir -p ~/.senpi/agent/extensions/subagent
-ln -sf "$(pwd)/packages/coding-agent/examples/extensions/subagent/index.ts" ~/.senpi/agent/extensions/subagent/index.ts
-ln -sf "$(pwd)/packages/coding-agent/examples/extensions/subagent/agents.ts" ~/.senpi/agent/extensions/subagent/agents.ts
-
-# Symlink agents
-mkdir -p ~/.senpi/agent/agents
-for f in packages/coding-agent/examples/extensions/subagent/agents/*.md; do
-  ln -sf "$(pwd)/$f" ~/.senpi/agent/agents/$(basename "$f")
-done
-
-# Symlink workflow prompts
-mkdir -p ~/.senpi/agent/prompts
-for f in packages/coding-agent/examples/extensions/subagent/prompts/*.md; do
-  ln -sf "$(pwd)/$f" ~/.senpi/agent/prompts/$(basename "$f")
-done
-```
-
-## Security Model
-
-This tool executes a separate `pi` subprocess with a delegated system prompt and tool/model configuration.
-
-**Project-local agents** (`.senpi/agents/*.md`) are repo-controlled prompts that can instruct the model to read files, run bash commands, etc.
-
-**Default behavior:** Only loads **user-level agents** from `~/.senpi/agent/agents`.
-
-To enable project-local agents, pass `agentScope: "both"` (or `"project"`). Only do this for repositories you trust.
-
-When running interactively, the tool prompts for confirmation before running project-local agents. Set `confirmProjectAgents: false` to disable.
-
-## Usage
-
-### Single agent
-```
-Use scout to find all authentication code
-```
-
-### Parallel execution
-```
-Run 2 scouts in parallel: one to find models, one to find providers
-```
-
-### Chained workflow
-```
-Use a chain: first have scout find the read tool, then have planner suggest improvements
-```
-
-### Workflow prompts
-```
-/implement add Redis caching to the session store
-/scout-and-plan refactor auth to support OAuth
-/implement-and-review add input validation to API endpoints
-```
-
-## Tool Modes
-
-| Mode | Parameter | Description |
-|------|-----------|-------------|
-| Single | `{ agent, task }` | One agent, one task |
-| Parallel | `{ tasks: [...] }` | Multiple agents run concurrently (max 8, 4 concurrent) |
-| Chain | `{ chain: [...] }` | Sequential with `{previous}` placeholder |
-
-## Output Display
-
-**Collapsed view** (default):
-- Status icon (✓/✗/⏳) and agent name
-- Last 5-10 items (tool calls and text)
-- Usage stats: `3 turns ↑input ↓output RcacheRead WcacheWrite $cost ctx:contextTokens model`
-
-**Expanded view** (Ctrl+O):
-- Full task text
-- All tool calls with formatted arguments
-- Final output rendered as Markdown
-- Per-task usage (for chain/parallel)
-
-**Parallel mode streaming**:
-- Shows all tasks with live status (⏳ running, ✓ done, ✗ failed)
-- Updates as each task makes progress
-- Shows "2/3 done, 1 running" status
-
-**Tool call formatting** (mimics built-in tools):
-- `$ command` for bash
-- `read ~/path:1-10` for read
-- `grep /pattern/ in ~/path` for grep
-- etc.
-
-## Agent Definitions
-
-Agents are markdown files with YAML frontmatter:
-
-```markdown
----
-name: my-agent
-description: What this agent does
-tools: read, grep, find, ls
-model: claude-haiku-4-5
----
-
-System prompt for the agent goes here.
-```
-
-**Locations:**
-- `~/.senpi/agent/agents/*.md` - User-level (always loaded)
-- `.senpi/agents/*.md` - Project-level (only with `agentScope: "project"` or `"both"`)
-
-Project agents override user agents with the same name when `agentScope: "both"`.
-
-## Sample Agents
-
-| Agent | Purpose | Model | Tools |
-|-------|---------|-------|-------|
-| `scout` | Fast codebase recon | Haiku | read, grep, find, ls, bash |
-| `planner` | Implementation plans | Sonnet | read, grep, find, ls |
-| `reviewer` | Code review | Sonnet | read, grep, find, ls, bash |
-| `worker` | General-purpose | Sonnet | (all default) |
-
-## Workflow Prompts
-
-| Prompt | Flow |
-|--------|------|
-| `/implement <query>` | scout → planner → worker |
-| `/scout-and-plan <query>` | scout → planner |
-| `/implement-and-review <query>` | worker → reviewer → worker |
-
-## Error Handling
-
-- **Exit code != 0**: Tool returns error with stderr/output
-- **stopReason "error"**: LLM error propagated with error message
-- **stopReason "aborted"**: User abort (Ctrl+C) kills subprocess, throws error
-- **Chain mode**: Stops at first failing step, reports which step failed
-
-## Limitations
-
-- Output truncated to last 10 items in collapsed view (expand to see all)
-- Agents discovered fresh on each invocation (allows editing mid-session)
-- Parallel mode limited to 8 tasks, 4 concurrent

package/examples/extensions/subagent/agents/planner.md

@@ -1,37 +0,0 @@
----
-name: planner
-description: Creates implementation plans from context and requirements
-tools: read, grep, find, ls
-model: claude-sonnet-4-5
----
-
-You are a planning specialist. You receive context (from a scout) and requirements, then produce a clear implementation plan.
-
-You must NOT make any changes. Only read, analyze, and plan.
-
-Input format you'll receive:
-- Context/findings from a scout agent
-- Original query or requirements
-
-Output format:
-
-## Goal
-One sentence summary of what needs to be done.
-
-## Plan
-Numbered steps, each small and actionable:
-1. Step one - specific file/function to modify
-2. Step two - what to add/change
-3. ...
-
-## Files to Modify
-- `path/to/file.ts` - what changes
-- `path/to/other.ts` - what changes
-
-## New Files (if any)
-- `path/to/new.ts` - purpose
-
-## Risks
-Anything to watch out for.
-
-Keep the plan concrete. The worker agent will execute it verbatim.

package/examples/extensions/subagent/agents/reviewer.md

@@ -1,35 +0,0 @@
----
-name: reviewer
-description: Code review specialist for quality and security analysis
-tools: read, grep, find, ls, bash
-model: claude-sonnet-4-5
----
-
-You are a senior code reviewer. Analyze code for quality, security, and maintainability.
-
-Bash is for read-only commands only: `git diff`, `git log`, `git show`. Do NOT modify files or run builds.
-Assume tool permissions are not perfectly enforceable; keep all bash usage strictly read-only.
-
-Strategy:
-1. Run `git diff` to see recent changes (if applicable)
-2. Read the modified files
-3. Check for bugs, security issues, code smells
-
-Output format:
-
-## Files Reviewed
-- `path/to/file.ts` (lines X-Y)
-
-## Critical (must fix)
-- `file.ts:42` - Issue description
-
-## Warnings (should fix)
-- `file.ts:100` - Issue description
-
-## Suggestions (consider)
-- `file.ts:150` - Improvement idea
-
-## Summary
-Overall assessment in 2-3 sentences.
-
-Be specific with file paths and line numbers.

package/examples/extensions/subagent/agents/scout.md

@@ -1,50 +0,0 @@
----
-name: scout
-description: Fast codebase recon that returns compressed context for handoff to other agents
-tools: read, grep, find, ls, bash
-model: claude-haiku-4-5
----
-
-You are a scout. Quickly investigate a codebase and return structured findings that another agent can use without re-reading everything.
-
-Your output will be passed to an agent who has NOT seen the files you explored.
-
-Thoroughness (infer from task, default medium):
-- Quick: Targeted lookups, key files only
-- Medium: Follow imports, read critical sections
-- Thorough: Trace all dependencies, check tests/types
-
-Strategy:
-1. grep/find to locate relevant code
-2. Read key sections (not entire files)
-3. Identify types, interfaces, key functions
-4. Note dependencies between files
-
-Output format:
-
-## Files Retrieved
-List with exact line ranges:
-1. `path/to/file.ts` (lines 10-50) - Description of what's here
-2. `path/to/other.ts` (lines 100-150) - Description
-3. ...
-
-## Key Code
-Critical types, interfaces, or functions:
-
-```typescript
-interface Example {
-  // actual code from the files
-}
-```
-
-```typescript
-function keyFunction() {
-  // actual implementation
-}
-```
-
-## Architecture
-Brief explanation of how the pieces connect.
-
-## Start Here
-Which file to look at first and why.

package/examples/extensions/subagent/agents/worker.md

@@ -1,24 +0,0 @@
----
-name: worker
-description: General-purpose subagent with full capabilities, isolated context
-model: claude-sonnet-4-5
----
-
-You are a worker agent with full capabilities. You operate in an isolated context window to handle delegated tasks without polluting the main conversation.
-
-Work autonomously to complete the assigned task. Use all available tools as needed.
-
-Output format when finished:
-
-## Completed
-What was done.
-
-## Files Changed
-- `path/to/file.ts` - what changed
-
-## Notes (if any)
-Anything the main agent should know.
-
-If handing off to another agent (e.g. reviewer), include:
-- Exact file paths changed
-- Key functions/types touched (short list)