🔐 pi-permission-system

Permission enforcement extension for the Pi coding agent that provides centralized, deterministic permission gates for tool, bash, MCP, skill, and special operations.

Coming from OpenCode?

Yes — this extension was designed so OpenCode-style agent permission policies can be ported into Pi with minimal friction.

Start here

| If you have this in OpenCode | In Pi, use this | |---|---| | Agent markdown file | ~/.pi/agent/agents/<agent-name>.md (respects PI_CODING_AGENT_DIR) | | YAML frontmatter | Same place: top of the markdown file | | Agent instructions / system prompt body | Same file, below frontmatter | | Agent permission rules | permission: inside that same frontmatter |

Important compatibility notes

• Agents are still markdown files with YAML frontmatter.
• Wildcard permissions still use last-match-wins ordering.
• Keep frontmatter simple when porting. This extension intentionally supports key: value scalars and nested maps, not full YAML features like arrays, anchors, or multiline scalars.

Minimal Pi agent example

---
name: my-agent
mode: primary
description: My ported agent
permission:
  tools:
    read: allow
    grep: allow
  bash:
    "*": ask
  mcp:
    "*": ask
---

Your agent instructions go here.

Compatibility matrix

| OpenCode concept | Pi equivalent with this extension | Compatibility | Porting notes | |---|---|---:|---| | Agent markdown files with YAML frontmatter | ~/.pi/agent/agents/<agent-name>.md | High | Your agent-local permission: frontmatter pattern carries over cleanly. | | Wildcard precedence | Same last-declared-match-wins behavior | High | Broad rules first, specific overrides later. | | bash permission rules | permission.bash | High | Command-pattern gating ports cleanly. | | Per-tool permission rules like read, grep, list, task, or arbitrary extension tool names | permission.tools | Medium-High | Pi groups registered tool names under tools, including built-ins and extension tools. | | external_directory | permission.special.external_directory | Medium | Same idea, different location. | | doom_loop | permission.special.doom_loop | Medium | Same idea, different location. | | skill permission rules | permission.skills | Medium | Same purpose, but Pi uses a dedicated plural skills section. | | MCP-related access | permission.mcp for proxy targets, permission.tools for direct registered tools | Medium | This is the biggest Pi-specific difference: proxy MCP targets and direct tool names are intentionally split. | | OpenCode-specific permissions like webfetch, websearch, question, lsp, todowrite | Usually extension-specific Pi tool names under permission.tools | Low-Medium | These do not have universal built-in one-to-one Pi names; map them to the actual registered tools available in your Pi setup. |

Most important difference

In OpenCode, many permission names live in one broad permission namespace. In Pi with this extension, there is a deliberate split:

| Use this when... | Put the rule here | |---|---| | You are targeting the registered mcp proxy tool and its internal server/tool targets | permission.mcp | | You are targeting an actual registered tool name, including direct extension tools like context7_*, github_*, or exa_* | permission.tools |

Fast porting guide

| If your OpenCode agent has... | In Pi, do this | |---|---| | permission.bash rules | Move them into permission.bash | | permission.external_directory | Move it to permission.special.external_directory | | permission.doom_loop | Move it to permission.special.doom_loop | | permission.skill rules | Move them to permission.skills | | Tool-ish permissions like read, grep, list, task, or third-party tool names | Put them in permission.tools | | MCP server/tool target logic | Put proxy-target rules in permission.mcp |

Practical takeaway

If you are coming from OpenCode, you usually do not need to rewrite your whole agent. In most cases, porting is just:

1. Keep the agent markdown/frontmatter structure.
2. Move OpenCode-style tool permissions into Pi's tools section.
3. Move external_directory and doom_loop into special.
4. Split MCP proxy target rules into mcp and direct registered tool rules into tools.

Features

• Tool Filtering — Hides disallowed tools from the agent before it starts (reduces "try another tool" behavior)
• System Prompt Sanitization — Removes denied tool entries from the Available tools: system prompt section so the agent only sees tools it can actually call
• Runtime Enforcement — Blocks/asks/allows at tool call time with UI confirmation dialogs and readable approval summaries
• Bash Command Control — Wildcard pattern matching for granular bash command permissions
• MCP Access Control — Server and tool-level permissions for MCP operations
• Skill Protection — Controls which skills can be loaded or read from disk, including multi-block prompt sanitization
• Per-Agent Overrides — Agent-specific permission policies via YAML frontmatter
• Subagent Permission Forwarding — Forwards ask confirmations from non-UI subagents back to the main interactive session
• Runtime YOLO Control — Lets users toggle yolo mode from the settings modal and lets other extensions toggle it through the runtime API
• File-Based Review Logging — Writes permission request/denial review entries to a file by default for later auditing
• Optional Debug Logging — Keeps verbose extension diagnostics in a separate file when enabled in config.json
• JSON Schema Validation — Full schema for editor autocomplete and config validation
• External Directory Guard — Enforces special.external_directory for path-bearing file tools that target paths outside the active working directory

Installation

npm package

pi install npm:pi-permission-system

Local extension folder

Place this folder in one of the following locations:

| Scope | Path | |---------|------| | Global default | ~/.pi/agent/extensions/pi-permission-system (respects PI_CODING_AGENT_DIR) | | Project | .pi/extensions/pi-permission-system |

Pi auto-discovers extensions in these paths.

Tip: All ~/.pi/agent paths shown in this document are defaults. If the PI_CODING_AGENT_DIR environment variable is set, pi uses that directory instead. The extension automatically follows pi's getAgentDir() helper for extension installation, session directories, and extension-local config paths. If you need policy lookup to come from a different global agent root, set PI_PERMISSION_SYSTEM_POLICY_AGENT_DIR.

Usage

Quick Start

1. Create the global policy file at the Pi agent runtime root (default: ~/.pi/agent/pi-permissions.jsonc, respects PI_CODING_AGENT_DIR):

{
  "defaultPolicy": {
    "tools": "ask",
    "bash": "ask",
    "mcp": "ask",
    "skills": "ask",
    "special": "ask"
  },
  "tools": {
    "read": "allow",
    "write": "deny"
  }
}

2. Start Pi — the extension automatically loads and enforces your policy.

Permission States

All permissions use one of three states:

| State | Behavior | |---------|---------------------------------------------| | allow | Permits the action silently | | deny | Blocks the action with an error message | | ask | Prompts the user for confirmation via UI |

Pi Integration Hooks

The extension integrates via Pi's lifecycle hooks:

| Hook | Behavior | |----------------------|-------------------------------------------------------------------------------------------| | before_agent_start | Filters active tools, removes denied tool entries from the system prompt, and hides denied skills | | tool_call | Enforces permissions for every tool invocation | | input | Intercepts /skill:<name> requests and enforces skill policy |

Additional behaviors:

• Unknown/unregistered tools are blocked before permission checks (prevents bypass attempts)
• The Available tools: system prompt section is rewritten to match the filtered active tool set
• Extension-provided tools like task, mcp, and third-party tools are handled through the same registered-tool permission layer instead of private built-in hardcodes
• When a subagent hits an ask permission without direct UI access, the request can be forwarded to the main interactive session for confirmation
• Generic extension-tool approval prompts include a bounded input preview; built-in file tools use concise human-readable summaries instead of raw multiline JSON
• Permission review logs include requested bash command text plus redacted prompt/input metadata for auditing without writing raw prompts or generic tool payload previews
• Path-bearing file tools (read, write, edit, find, grep, ls) evaluate special.external_directory before their normal tool permission when an explicit path points outside ctx.cwd

Configuration

Extension Config File

Location: global Pi extension config (default: ~/.pi/agent/extensions/pi-permission-system/config.json, respects PI_CODING_AGENT_DIR)

Set PI_PERMISSION_SYSTEM_CONFIG_PATH to point this extension at a specific config file when the default global path is not appropriate.

The extension creates this file automatically when it is missing. It controls extension-local logging behavior and yolo mode defaults:

{
  "debugLog": false,
  "permissionReviewLog": true,
  "yoloMode": false
}

| Key | Default | Description | |-----|---------|-------------| | debugLog | false | Enables verbose diagnostic logging to logs/pi-permission-system-debug.jsonl | | permissionReviewLog | true | Enables the permission request/denial review log at logs/pi-permission-system-permission-review.jsonl | | yoloMode | false | Auto-approves ask results instead of prompting when yolo mode is enabled |

Both logs write to files only under the extension directory by default. Set PI_PERMISSION_SYSTEM_LOGS_DIR to redirect review/debug logs to a specific directory. No debug output is printed to the terminal.

Runtime YOLO Control

Use /permission-system to open the settings modal and inspect or change yolo mode interactively.

Other extensions can toggle yolo mode immediately through the shared runtime API:

type PermissionSystemGlobal = typeof globalThis & {
  __piPermissionSystem?: {
    toggleYoloMode(options?: { persist?: boolean; source?: string }): { error?: string };
  };
};

pi.registerShortcut("f8", {
  description: "Toggle pi-permission-system YOLO mode",
  handler: () => {
    const permissionSystem = (globalThis as PermissionSystemGlobal).__piPermissionSystem;
    const result = permissionSystem?.toggleYoloMode({ source: "my-extension" });
    if (result?.error) {
      // Notify or log the error in your extension.
    }
  },
});

The runtime API exposes getYoloMode(), setYoloMode(enabled, options?), and toggleYoloMode(options?). Runtime updates persist to config.json by default; pass { persist: false } for a current-session-only toggle.

Global Policy File

Location: global Pi policy file (default: ~/.pi/agent/pi-permissions.jsonc, respects PI_PERMISSION_SYSTEM_POLICY_AGENT_DIR when set and otherwise follows PI_CODING_AGENT_DIR)

The policy file is a JSON object with these sections:

| Section | Description | |-----------------|-----------------------------------------------------| | defaultPolicy | Fallback permissions per category | | tools | Pattern-based tool permissions for registered tools | | bash | Command pattern permissions | | mcp | MCP server/tool permissions for calls routed through a registered mcp tool | | skills | Skill name pattern permissions | | special | Reserved permission checks such as external directory access |

Note: JSONC comments and trailing commas are supported. If parsing still fails, the extension falls back to ask for all categories and shows a warning in the TUI when available.

Global Per-Agent Overrides

Override global permissions for specific agents via YAML frontmatter in the global Pi agents directory (default: ~/.pi/agent/agents/<agent-name>.md, respects PI_PERMISSION_SYSTEM_POLICY_AGENT_DIR when set and otherwise follows PI_CODING_AGENT_DIR):

---
name: my-agent
permission:
  tools:
    read: allow
    write: deny
    mcp: allow
  bash:
    git status: allow
    git *: ask
  mcp:
    chrome_devtools_*: deny
    exa_*: allow
  skills:
    "*": ask
---

MCP behavior: permission.tools.mcp is the coarse entry/fallback permission for a registered mcp tool when one is available. More specific permission.mcp target rules override that fallback when they match.

Limitations: The frontmatter parser is intentionally minimal. Use only key: value scalars and nested maps. Avoid arrays, multi-line scalars, and YAML anchors. If you are porting from OpenCode, simplify richer YAML frontmatter before expecting a clean migration.

Project-Level Policy Files

The extension can also layer project-local permission files relative to the active session working directory:

| Scope | Path | |-------|------| | Project policy | <cwd>/.pi/agent/pi-permissions.jsonc | | Project agent override | <cwd>/.pi/agent/agents/<agent-name>.md |

Project-local files use the same formats as the global policy file and global agent frontmatter. These project files are resolved from Pi's current session cwd, so they are workspace-specific and do not move under PI_CODING_AGENT_DIR.

Precedence order:

1. Global policy file
2. Project policy file
3. Global agent frontmatter
4. Project agent frontmatter

Later trusted layers override earlier layers within the same permission category, and project-local layers can tighten policy by adding deny rules. Project-local policy cannot relax a deny from the global policy file or global agent frontmatter: an allow or ask in a project policy is ignored when the latest matching trusted layer is deny. For wildcard-based sections like tools, bash, mcp, skills, and special, matching still follows last matching rule wins within the applicable trust boundary, with global/system deny rules acting as floors for project-local overrides.

Policy Reference

defaultPolicy

Sets fallback permissions when no specific rule matches:

{
  "defaultPolicy": {
    "tools": "ask",
    "bash": "ask",
    "mcp": "ask",
    "skills": "ask",
    "special": "ask"
  }
}

tools

Controls tools by registered name pattern. This is the recommended standalone format for all tool entries, including Pi built-ins and arbitrary third-party extension tools. Patterns use * wildcards and follow last-declared-match semantics, so put broad fallbacks first and specific overrides later.

| Tool name example | Description | |-----------------------|-------------| | bash | Shell command execution (tool-level fallback before bash pattern rules) | | read / write | Canonical Pi built-in file tools | | mcp | Registered MCP proxy tool entry/fallback when available | | task | Delegation tool handled like any other registered extension tool | | third_party_tool | Arbitrary registered extension tool | | context7_* | Wildcard for direct tools registered by another extension | | * | Fallback for every registered tool not matched by a later rule |

{
  "tools": {
    "*": "ask",
    "context7_*": "ask",
    "third_party_tool": "ask",
    "mcp": "allow",
    "read": "allow",
    "write": "deny"
  }
}

Unknown or absent tools are not required in the config. If another extension is not installed, its tool simply will not be registered at runtime, and this extension will block attempts to call that missing tool before permission checks run. Wildcard tools rules apply to direct tools from any extension; no adapter-specific naming is required.

Note: Setting tools.bash affects the default for bash commands, but bash patterns can provide command-level overrides.

Note: Setting tools.mcp controls coarse access to a registered mcp proxy tool when one is available. Specific mcp rules still override it when a proxy target pattern matches. Direct MCP tools registered by extensions are regular registered tools and should be controlled with tools patterns such as context7_* or github_*.

Note: Top-level shorthand is only supported for the canonical Pi built-ins (bash, read, write, edit, grep, find, ls) in agent frontmatter. Use permission.tools.<name> for mcp, task, and any third-party tool.

bash

Command patterns use * wildcards and match against the full command string. If multiple patterns match, the last declared matching rule wins. Put broad fallback rules first and more specific overrides later.

{
  "bash": {
    "git *": "ask",
    "git status": "allow",
    "rm -rf *": "deny"
  }
}

mcp

MCP permissions match against derived targets from tool input. These rules are more specific than tools.mcp and override that fallback when a pattern matches:

| Target Type | Examples | |-------------------|---------------------------------------------| | Baseline ops | mcp_status, mcp_list, mcp_search, mcp_describe, mcp_connect | | Server name | myServer | | Server/tool combo | myServer:search, myServer_search | | Generic | mcp_call |

{
  "mcp": {
    "*": "ask",
    "myServer:*": "ask",
    "mcp_status": "allow",
    "mcp_list": "allow",
    "dangerousServer": "deny"
  }
}

Note: Baseline discovery targets may auto-allow when you permit any MCP rule.

MCP Tool Fallback via tools.mcp

A registered mcp tool can use tools.mcp as an entry permission point. This provides a fallback when no specific MCP pattern matches:

{
  "tools": {
    "mcp": "allow"
  }
}

This is useful for per-agent configurations where you want to grant MCP access broadly:

# In the global Pi agents directory (default: ~/.pi/agent/agents/researcher.md; respects PI_CODING_AGENT_DIR)
---
name: researcher
permission:
  tools:
    mcp: allow
---

The permission resolution order for MCP operations:

1. Specific mcp patterns (e.g., myServer:toolName, myServer_*)
2. tools.mcp fallback (if set)
3. defaultPolicy.mcp

skills

Skill name patterns use * wildcards:

{
  "skills": {
    "*": "ask",
    "dangerous-*": "deny"
  }
}

special

Reserved permission checks:

| Key | Description | |----------------------|------------------------------------------| | doom_loop | Controls doom loop detection behavior | | external_directory | Enforces ask/allow/deny decisions for path-bearing built-in tools (read, write, edit, find, grep, ls) when they target paths outside the active working directory |

{
  "special": {
    "doom_loop": "deny",
    "external_directory": "ask"
  }
}

external_directory is evaluated before the normal tool permission check. For example, tools.read: "allow" can permit ordinary reads while special.external_directory: "ask" still requires confirmation before reading ../outside.txt or an absolute path outside ctx.cwd. Optional-path search tools (find, grep, ls) skip this check when no path is provided because they default to the active working directory.

Common Recipes

Read-Only Mode

{
  "defaultPolicy": { "tools": "ask", "bash": "ask", "mcp": "ask", "skills": "ask", "special": "ask" },
  "tools": {
    "read": "allow",
    "grep": "allow",
    "find": "allow",
    "ls": "allow",
    "write": "deny",
    "edit": "deny"
  }
}

Restricted Bash Surface

{
  "defaultPolicy": { "tools": "ask", "bash": "deny", "mcp": "ask", "skills": "ask", "special": "ask" },
  "bash": {
    "git *": "ask",
    "git status": "allow",
    "git diff": "allow",
    "git log *": "allow"
  }
}

MCP Discovery Only

{
  "defaultPolicy": { "tools": "ask", "bash": "ask", "mcp": "ask", "skills": "ask", "special": "ask" },
  "mcp": {
    "*": "ask",
    "mcp_status": "allow",
    "mcp_list": "allow",
    "mcp_search": "allow",
    "mcp_describe": "allow"
  }
}

Per-Agent Lockdown

In the global Pi agents directory (default: ~/.pi/agent/agents/reviewer.md, respects PI_CODING_AGENT_DIR):

---
permission:
  tools:
    write: deny
    edit: deny
  bash:
    "*": deny
---

Technical Details

Permission Prompt Summaries

When a tool permission resolves to ask, the prompt is designed to be readable enough for an informed approval decision:

• bash prompts show the command and matched bash pattern when available.
• mcp prompts show the derived MCP target and matched rule when available.
• Built-in file tools show concise summaries, such as the target path and edit/write line counts, instead of raw multiline JSON.
• Unknown or third-party extension tools show a bounded single-line JSON preview of the input so users are not asked to approve a blind tool name.

Example edit approval prompt:

Current agent requested tool 'edit' for '.gitignore' (1 replacement: edit #1 replaces 5 lines with 2 lines). Allow this call?

Subagent Permission Forwarding

When a delegated or routed subagent runs without direct UI access, ask permissions can still be enforced by forwarding the confirmation request through Pi session directories. The main interactive session polls for forwarded requests, shows the confirmation prompt, writes the response, and the subagent resumes once that decision is available.

This keeps ask policies usable even when the original permission check happens inside a non-UI execution context.

Logging

When the extension prompts, denies, or forwards permission requests, it can append structured JSONL entries under:

Default global logs directory: ~/.pi/agent/extensions/pi-permission-system/logs/
Actual global logs directory: $PI_CODING_AGENT_DIR/extensions/pi-permission-system/logs when PI_CODING_AGENT_DIR is set
Override logs directory: $PI_PERMISSION_SYSTEM_LOGS_DIR when set

• pi-permission-system-permission-review.jsonl — enabled by default for permission review/audit history, including metadata hashes and lengths for prompts, commands, denial reasons, and tool input previews instead of raw sensitive content
• pi-permission-system-debug.jsonl — disabled by default and intended for troubleshooting

Architecture

index.ts                         → Root Pi entrypoint shim
src/
├── index.ts                     → Extension bootstrap, permission checks, readable prompts, review logging, reload handling, and subagent forwarding
├── before-agent-start-cache.ts  → Caches prompt/tool filtering state between before_agent_start runs
├── bash-filter.ts               → Bash command wildcard pattern matching
├── common.ts                    → Shared utilities (YAML parsing, type guards, etc.)
├── config-modal.ts              → `/permission-system` modal registration and settings UI wiring
├── extension-config.ts          → Extension-local config loading and default creation
├── logging.ts                   → File-only debug/review logging helpers
├── model-option-compatibility.ts → Guards unsupported provider/model options
├── permission-dialog.ts         → Interactive permission approval UI helpers
├── permission-forwarding.ts     → Subagent-to-parent permission forwarding utilities
├── permission-manager.ts        → Global/project policy loading, merging, and resolution with caching
├── skill-prompt-sanitizer.ts    → Skill prompt parsing, multi-block sanitization, and skill-read path matching
├── status.ts                    → Status line integration for runtime yolo state
├── system-prompt-sanitizer.ts   → Available-tools prompt filtering helpers
├── tool-registry.ts             → Registered tool name resolution
├── types.ts                     → TypeScript type definitions
├── wildcard-matcher.ts          → Shared wildcard pattern compilation and matching
├── yolo-mode.ts                 → Runtime yolo approval helpers
├── yolo-mode-api.ts             → Shared global runtime API for yolo toggling
└── zellij-modal.ts              → Reusable modal/settings UI components
tests/
├── permission-system.test.ts    → Core permission, layering, forwarding, and policy tests
├── config-modal.test.ts         → Modal command behavior tests
└── test-harness.ts              → Shared lightweight test helpers
schemas/
└── permissions.schema.json      → JSON Schema for policy validation
config/
└── config.example.json          → Starter global policy template

Module Organization

The extension uses a modular architecture with shared utilities:

| Module | Purpose | |--------|---------| | common.ts | Shared utilities: toRecord(), getNonEmptyString(), isPermissionState(), parseSimpleYamlMap(), extractFrontmatter() | | wildcard-matcher.ts | Compile-once wildcard patterns with specificity sorting: compileWildcardPatterns(), findCompiledWildcardMatch() | | permission-manager.ts | Policy resolution with file stamp caching for performance | | bash-filter.ts | Uses shared wildcard matcher for bash command patterns | | skill-prompt-sanitizer.ts | Parses all available skill prompt blocks, removes denied skills, and tracks visible skill paths for read protection |

Performance Optimizations

• File stamp caching: Configurations are cached with file modification timestamps to avoid redundant reads
• Pre-compiled patterns: Wildcard patterns are compiled to regex once and reused across permission checks
• Resolved permissions caching: Merged agent+global permissions are cached per-agent with invalidation on file changes

Threat Model

Goal: Enforce policy at the host level, not the model level.

What this stops:

• Agent calling tools it shouldn't use (e.g., write, dangerous bash)
• Tool switching attempts (calling non-existent tool names)
• Accidental escalation via skill loading
• Unapproved path-bearing tool access outside the active working directory when external_directory is ask or deny

Limitations:

• If a dangerous action is possible via an allowed tool, policy must explicitly restrict it
• This is a permission decision layer, not a sandbox

Schema Validation

Validate your config against the included schema:

npx --yes ajv-cli@5 validate \
  -s ./schemas/permissions.schema.json \
  -d ./pi-permissions.valid.json

Editor tip: Add "$schema": "./schemas/permissions.schema.json" to your config for autocomplete support.

Troubleshooting

| Problem | Cause | Solution | |---------|-------|----------| | Config not applied (everything asks) | File not found or parse error | Verify the global Pi policy file (default: ~/.pi/agent/pi-permissions.jsonc, respects PI_CODING_AGENT_DIR); check the TUI warning for the parse location/message | | Per-agent override not applied | Frontmatter parsing issue | Ensure --- delimiters at file top; keep YAML simple; restart session | | Tool blocked as unregistered | Unknown tool name | Use a registered mcp tool for server tools: { "tool": "server:tool" } | | /skill:<name> blocked | Deny policy or confirmation unavailable | Check merged skills policy (global/project/agent layers). Active agent context is optional in the main session; ask still requires UI or forwarded confirmation. | | External file path blocked | special.external_directory is ask without UI or deny | Allow/ask the special permission or keep file tools inside the active working directory. | | Permission prompt is too verbose | Generic extension tool input is large | Built-in file tools are summarized automatically; third-party tools are capped to a bounded one-line JSON preview. |

Development

Runtime checks require Node.js 20+; the test suite requires Bun 1.1+.

npm run build              # Run TypeScript type checks
npm run lint               # Run local static checks
npm run validate:artifacts # Validate JSON/schema/example artifacts
npm run test               # Run Bun tests from ./tests
npm run check              # Run static, artifact, and test checks

Related Pi Extensions

• pi-multi-auth — Multi-provider credential management and quota-aware rotation
• pi-tool-display — Compact tool rendering and diff visualization
• pi-rtk-optimizer — RTK command rewriting and output compaction
• pi-MUST-have-extension — RFC 2119 keyword normalization for prompts

License

MIT