Pi Prompt Authoring

Pi Prompt Authoring

Authors custom system prompt additions for pi coding agent. Pi provides a default system prompt with identity, tools, and guidelines. This skill covers what to add via SYSTEM.md (replace default) or APPEND_SYSTEM.md (extend default).

How Pi's System Prompt Works

Pi builds the system prompt dynamically at runtime. The construction follows this order:

Default System Prompt Structure

When using the default prompt, pi constructs it with these sections:

1. Identity statement - "You are an expert coding assistant operating inside pi, a coding agent harness. You help users by reading files, executing commands, editing code, and writing new files."
2. Available tools - List of enabled tools with one-line descriptions
3. Guidelines - Context-aware guidelines based on which tools are available
4. Pi documentation references - Paths to README, docs, and examples
5. Append system prompt - Content from APPEND_SYSTEM.md or --append-system-prompt
6. Project context - Loaded AGENTS.md/CLAUDE.md files
7. Skills section - XML-formatted list of available skills (only if read tool is enabled)
8. Metadata - Current date and working directory (always last)

Tool Listing

Tools appear in the system prompt only when a one-line description snippet is provided. Default tools are read, bash, edit, write. The format is:

Available tools:
- read: Read the contents of a file
- bash: Execute a bash command in the current working directory
- edit: Edit a single file using exact text replacement
- write: Write content to a file

Custom tools from extensions can also be included if the extension provides a snippet. If no tools have snippets, the prompt shows "(none)".

Guidelines

Guidelines are context-aware based on available tools:

• If bash is available without grep/find/ls: "Use bash for file operations like ls, rg, find"
• If bash is available with grep/find/ls: "Prefer grep/find/ls tools over bash for file exploration (faster, respects .gitignore)"
• Always included: "Be concise in your responses" and "Show file paths clearly when working with files"

Custom guidelines can be added via extensions and appear before the default guidelines.

Skills Section Format

At startup, pi scans skill locations and extracts names, descriptions, and file paths. The system prompt includes visible skills in XML format per the Agent Skills specification:

The following skills provide specialized instructions for specific tasks.
Use the read tool to load a skill's file when the task matches its description.
When a skill file references a relative path, resolve it against the skill directory (parent of SKILL.md / dirname of the path) and use that absolute path in tool commands.

<available_skills>
  <skill>
    <name>skill-name</name>
    <description>What this skill does and when to use it</description>
    <location>/absolute/path/to/skill/SKILL.md</location>
  </skill>
</available_skills>

This is progressive disclosure: only the skill list is always in context; full skill instructions load on-demand when the agent uses read to load the SKILL.md file. Skills with disable-model-invocation: true in frontmatter are excluded from the prompt and can only be invoked via /skill:name commands.

Context File Loading Order

Pi loads AGENTS.md (or CLAUDE.md, in that priority order) from:

1. ~/.pi/agent/AGENTS.md (global, loaded first)
2. Parent directories walking up from cwd to filesystem root
3. Current directory (loaded last)

All matching files are concatenated in this order with headers:

# Project Context

Project-specific instructions and guidelines:

## /path/to/AGENTS.md

[content]

If both AGENTS.md and CLAUDE.md exist in the same directory, only AGENTS.md is used.

Custom System Prompt Behavior

SYSTEM.md or --system-prompt (replaces default):

• Your custom prompt becomes the base
• APPEND_SYSTEM.md or --append-system-prompt is still appended
• Context files (AGENTS.md/CLAUDE.md) are appended after
• Skills section is appended if read tool is available
• Date and working directory are always added last

APPEND_SYSTEM.md or --append-system-prompt (extends default):

• Appended to the default system prompt after documentation references
• Before context files and skills section

Custom System Prompt Locations

• Project: .pi/SYSTEM.md or .pi/APPEND_SYSTEM.md
• Global: ~/.pi/agent/SYSTEM.md or ~/.pi/agent/APPEND_SYSTEM.md
• CLI: --system-prompt <text> (replaces) or --append-system-prompt <text> (extends)

Project files take precedence over global files. CLI flags override file-based discovery.

Key principle: Don't duplicate what pi already provides. Focus your custom prompt on additional principles and constraints specific to your workflow.

Section 1: Core Principles (Optional)

Purpose: Define coding standards, code quality expectations, and architectural preferences. Pi's default is minimal; add principles for your workflow.

Common principle categories:

• Simplicity: Prefer simple solutions, abstraction is earned
• Code quality: No AI-generated debt, same standards as human code
• Codebase health: Delete dead code, fix issues on contact
• Working code: Finish what works, don't rewrite unnecessarily
• Build & verification: Code must build/lint/typecheck before shipping
• Security: Validate input, never log secrets, use allowlists

Example:

<simplicity>
The assistant uses the simplest code that solves the problem. Abstraction is earned.
</simplicity>

<codebase_health>
Every change must leave the codebase healthier. Delete dead code.
</codebase_health>

Section 2: Behavioral Guidelines (Optional)

Purpose: Enforce specific behaviors beyond pi's defaults. Define how the agent should approach tasks, handle errors, and interact with users.

Common guideline categories:

• Debugging: Fix causes not symptoms, read implementation before changing
• Architecture: Dependencies flow one direction, interfaces belong to consumers
• Testing: Tests verify what code does, each test must earn its place
• Scope: One change does one thing, explicit permission boundaries
• Error handling: Fail fast, no fallback defaults masking errors
• Documentation: Read docs before configuring, use upstream examples
• User directives: User corrections are permanent facts, repeated requests mean failure

Example:

<debugging_and_fixes>
Fix the cause, not the symptom. Read the implementation, trace the box model,
understand the pixels, then change one thing with certainty.
</debugging_and_fixes>

<scope>
One change does one thing. No unrequested features, no undiscussed removals.
</scope>

Section 3: Custom Tool Usage Guidelines (Optional)

Purpose: Add usage patterns for custom tools registered via pi extensions. Pi auto-generates tool descriptions for built-in tools; only add guidelines for extension-specific tools.

Example for custom display tool:

<display_requirements>
The display must show what you are doing. Update at phase changes.
Use genui tool with openui-lang source for display updates.
</display_requirements>

<genui_tool>
Use Canvas with Header, Col, and Timestamp. Header includes icon, title, subtitle.
Icons by phase: sync for running, check for done, bug for error.
</genui_tool>

Section 4: Project-Specific Context (Optional)

Purpose: Add project-specific conventions, commands, or constraints. Pi already loads AGENTS.md/CLAUDE.md files; use APPEND_SYSTEM.md for rules that don't fit there.

Example:

<project_context>
The project uses Nix flakes. Always use `nix develop` for shell access.
Build with `nix build`. Tests run via `nix flake check`.
</project_context>

Architectural Patterns

Use APPEND_SYSTEM.md for extensions: Don't replace pi's default unless necessary. Append custom principles to preserve built-in functionality.

Use SYSTEM.md for replacements: Only if you need to fundamentally change pi's behavior.

XML tags for organization: Use <section_name> tags to organize principles logically.

Concise, value-based rules: "Scope is sacred" not "When asked to do X, only touch files that Y".

No procedural details: Avoid code snippets, variable names, or step-by-step scenarios.

Validation Checklist

Before finalizing:

• No duplication of pi's default (identity, date, tools, skills)
• Principles are concise and value-based, not procedural
• XML tags used for logical organization
• No code snippets or variable names in principles
• Custom tool guidelines only for extension-specific tools
• Project context only for non-obvious conventions
• File placed correctly: SYSTEM.md (replace) or APPEND_SYSTEM.md (extend)
• No political, ethical, or moral points — these are not relevant and should not be in system prompts
• Focus on agent accountability and factuality — these are the only behavioral values that belong in system prompts