bosun

Bosun configuration — models, sandbox, daemon, Pi settings. Use when changing models, editing config, or understanding how bosun is set up.

Bootstrap a new project using bosun as a foundation — via bun dependency (recommended) or git submodule. Creates a downstream project that inherits bosun's multi-agent infrastructure while adding custom agents, skills, and extensions. Works standalone — fetchable by any pi agent via raw GitHub URL.

Search curated markdown memory like sessions, plans, docs, and skills. Use when recalling prior context or looking for relevant historical/project knowledge.

Analyze Pi session JSONL files using jq patterns. Use when extracting metrics, tool usage, costs, or reviewing session history. Load for session export, summarization, or workflow analysis.

Session-based browser plan review primitives for Bosun. Provides a local-first state model, markdown-aware anchor extraction, and review-session persistence for reviewing existing plan files in dedicated CDP browser windows. Triggers: "plan-review", "review this plan", "review markdown plan".

Session-based browser diff review primitives for Bosun. Provides transport- agnostic v1 schemas plus local state and git snapshot helpers for immutable review rounds. Triggers: "diff-review", "reround", "review round", "snapshot-backed review".

Browser automation via Chrome DevTools Protocol. Connects to Chromium running with --remote-debugging-port=9222. Use for: navigate, click, fill, screenshot, inspect DOM, capture console logs, debug errors, monitor network, viewport emulation, visual review, overflow detection. Triggers: "browser", "go to", "click", "fill form", "take screenshot", "web page", "scrape", "console errors", "debug", "network requests", "visual review", "responsive", "viewport", "overflow".

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, and filler phrases.

Orchestrate multiple Pi agents via tmux windows. Use when spawning background agents, sending them tasks, and capturing their output.

Multi-agent coordination via pi-mesh. Use when multiple agents work in the same project - reservations, messaging, and peer awareness.

Create Pi agent definitions with proper frontmatter, model tiers, and tool permissions. Use when building specialized agents for code review, documentation, security, research, etc.

Working with bosun's daemon system for automated workflows. Covers workflow directories, agent spawning, validators, and file-based chaining.

Generate images using Gemini's image generation API (Nano Banana models). Use when generating blog images, illustrations, diagrams, or any visual content. Supports gemini-2.5-flash-image, Nano Banana Pro, and Nano Banana 2.

Browser-to-mesh annotation bridge via CDP. Injects an annotator UI into a browser tab — select text, comment, and send structured annotations to any mesh agent. Agent responses appear as toasts in the browser. Uses CDP Runtime.addBinding for the communication channel (no HTTP/CORS/CSP issues). Triggers: "browser-bridge", "annotate", "browser annotation", "page review".

Manage plans, handoffs, forks, and pickups. Load for /handoff, /fork, /pickup, or when planning complex tasks.

Project management skill for Q agent. Provides qp CLI for project CRUD, task aggregation, and progress tracking. Use when managing projects or viewing task rollups.

Roadmap planning skill for Q agent. Provides qr CLI for roadmap CRUD and project aggregation. Use when planning quarterly roadmaps or coordinating across teams.

Task tracking skill for Q agent. Provides qt CLI for task CRUD, queries, dependencies, and archival. Use when managing tasks, tracking blockers, or triaging inbox.

When and how to load skills proactively. Use when starting any task to avoid missing conventions, especially for git, file operations, or project-specific work.

Generate public builder's logs from coding sessions. Analyzes plans and sessions, detects deviations, synthesizes narratives per journey. Use with /chronicle command.

Use when starting dev servers, long-running processes, or any command that shouldn't block. Covers createBackgroundProcess patterns and common mistakes.

Editorial review for written content (blog posts, documentation, READMEs). Use when reviewing prose for clarity, accuracy, voice, and structure. Complements code review - invoke via @review agent for written content.

Proactively load this skill for ANY git-related operation including cloning, repo organization, worktrees, branching, commits, merging, rebasing, and troubleshooting. Use when working with git in any capacity - cloning repos, creating worktrees, writing commits, resolving conflicts, or any other git workflow.

Use when working with GitHub via CLI - pull requests, issues, code reviews, or repository operations. Also load for first-time gh CLI authentication setup.

Create Pi prompt templates (slash commands) for repetitive tasks. Use when adding /commands that trigger specific prompts or workflows.

Create Pi extensions with event handling, commands, and hooks. Use when building extensions that respond to session events or add custom commands.

Create new agent skills following the Agent Skills specification. Use when building reusable skills with proper structure and frontmatter.

Create custom Pi tools via extensions that the LLM can call. Use when adding new callable functions for the LLM.

Create Pi daemon workflows with proper structure, config, agents, and validators. Use when building automated workflows for the daemon system.

Review and sync skill for Q agent. Monitors changes across tasks/projects/roadmaps, maintains update logs, and ensures consistency. Use after bulk changes or during standups.