Finalise AFK loop worktree after smoke/refine
Focussed instruction set for scripted AFK loop
Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.
Human-in-the-loop single tick for AFK-style task work
hand over decision making
Pi-specific agent tooling extracted from my dotfiles repo. Still need to port more of the old claude stuff.
pi/extensions/ — package-shipped Pi extensions grouped into tools/, ui/, messaging/, cli/, and system-prompt/.pi/extensions/ — project-local Pi extensions for this repository onlypi/agents/ — bundled subagents used by the subagent extensionplugins/ — reusable prompt/skill plugins grouped by domain (devflow/, writing/, coding/, harness/)pi/themes/ — Pi themesNote on Cache Invalidation: Changing prompt-layer context mid-session, or changing the model/provider, can drop provider prompt-cache reuse. Treat dynamic context injection as a cost/latency tradeoff; keep injected context bounded and stable when possible.
From a local checkout:
pi install /absolute/path/to/agents
From git later:
pi install git:github.com/<you>/agents
Pi loads the package's extensions from pi/extensions/, plugin prompts and skills from plugins/*/, themes from pi/themes/, through package.json#pi. It also auto-loads this checkout's project-local extensions from .pi/extensions/ when Pi runs inside this repository.
The bundled agents are discovered by the subagent extension from pi/agents/, so they travel with the package too. Project-specific agents still load from the nearest .pi/agents/ directory when you run Pi inside another repo.
This package ships a system-prompt extension that:
subagent toolagent.njk rulesIt also ships a project-structure messaging extension that sends a bounded project tree as model-visible custom message context.
This repository additionally keeps a project-local .pi/extensions/pi-internals/ tool that agents can call on demand to print Pi runtime/source/settings/enabled-extension paths. It is intentionally local to this checkout rather than shipped as part of the package.
See pi/extensions/system-prompt/README.md for prompt-layer extension details and .pi/extensions/pi-internals/README.md for project-local Pi internals discovery.
pi/agents/scout.md — fast codebase recon and architecture mappingpi/agents/fixer.md — validation repair and scoped mechanical completion agentpi/agents/hack.md — shell-first investigation and automation agentpi/agents/review.md — faster/cheaper single-agent code review for spot checkspi/agents/deep-review/ — multi-role review for a full feature, PR, or session workload; pass commits/diff, PRD/spec/task files, relevant paths, intent, risks, validation results, and any upfront exploration notes so each reviewer starts with shared hot contextpi/agents/council/ — multi-role ideation panel for non-trivial decisions where the main agent wants another opinion before reporting back; includes skeptic, evidence scout, simplifier, and scope guard; pass proposed direction, intended outcome, problem, tradeoff/decision, relevant files/specs/code paths, constraints, risks, rejected options, and desired helpplugins/devflow/ — AFK workflow prompts/skill plus HITL, shrug, and grill-me promptsplugins/writing/ — skill authoring, spec authoring, and Mermaid writing skillsplugins/coding/ — git, robustness, writing-tests skills plus coding promptsplugins/harness/ — harness/session affordances such as Pi session introspection and rich responsesThe Vitest suite includes both unit/snapshot tests and Pi runtime integration tests backed by @gaodes/pi-test-harness.
pnpm install
pnpm format
pnpm lint
pnpm typecheck
pnpm check
Running pi from this repo works for local testing because .pi/settings.json points Pi at the package root.
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub codethread/agents --plugin devflowPhase-based development workflow: What, How, Build. Structured planning with research, learning tests, and prototyping before implementation.
Pithos hooks for Claude Code. Tracks runs/tasks/artifacts for spawner-launched sessions.
Scope gatekeeper agent that pushes back on unnecessary features and complexity. Delegate to Karen before implementing new features to challenge necessity, detect optimistic bias, and insist on the simplest solution.
Opinionated project architecture bootstrapping and migration
Push notifications for Claude Code via cc-notify server (Stop, PermissionRequest, activity tracking)
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.
Multi-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Write feature specs, plan roadmaps, and synthesize user research faster. Keep stakeholders updated and stay ahead of the competitive landscape.