my-pi-again

Skill chain resolver extension. Resolves `imports:` and `Read skill:` references in SKILL.md files recursively, making chained skills available to pi as first-class skills. Use /skill-chain to inspect and manage skill chains.

Independent execution pipeline. Takes any plan markdown as input, breaks it into discrete tasks with dependencies, and orchestrates specialized sub-agents to implement them. Enforces 0 lint/build errors after every task. Resumable if interrupted. Use when the user says "deep execute", "/deep-execute", or wants to implement a plan.

Launch a multi-phase planning pipeline: research the codebase AND web for best practices, create a plan, generate a user flow spec, deepen it, get critiques from specialized agents, resolve questions, and produce a finalized plan. All state saves to docs/plans/{slug}/ as committable markdown. Resumable if interrupted. Use when the user says "deep plan", "/deep-plan", or asks for a thorough implementation plan.

Protocol for preserving critical skill invariants across context compaction. Imported by orchestrator skills (deep-plan, deep-execute, deep-review, deep-auto). Uses PreCompact + PostCompact hooks to ensure non-negotiable rules survive context loss.

Fully autonomous pipeline orchestrator. Chains deep-plan (with auto-resolved Q&A), deep-execute, deep-review, and optionally deep-complete into a single unattended run. All critic questions are resolved using codebase evidence and codified best practices. Resumable if interrupted. Use when the user says "deep auto", "/deep-auto", or wants to run the full pipeline autonomously.

Structured bug-fix pipeline. Reproduce → validate intent → write red test → fix → verify → regression review → complete. Never fixes without a failing test first. Resumable. Use when the user says "deep bug", "/deep-bug", or provides a bug report.

Completion phase. Commits code, generates documentation, and produces a final summary. Run after deep-review passes. Use when the user says "deep complete", "/deep-complete", or wants to finalize the pipeline.

Post-pipeline fix-up skill. Root cause analysis for every issue, TDD fixes, and spec/doc amendments. Never patches without understanding why. Resumable.

Self-improving skill that analyzes pipeline history, feedback, and recurring issues to propose concrete improvements to skills and agents. Runs periodically.

Self-improving skill that analyzes pipeline failures and user frustration, traces them to skill/agent gaps, and proposes concrete amendments. Fixes the immediate problem first, then reflects.

Deep internet research pipeline. Spawns parallel research agents across multiple angles (market, UX, pain points, trends), synthesizes findings, and produces actionable insights. Can run standalone or integrated into deep-plan. Use when the user says "deep research" or `/deep-research`.

Multi-round review pipeline. Spawns blind reviewers (quality, security, a11y, performance, data integrity, completeness), runs the full test suite, triages findings, fixes them, and verifies. Runs N rounds until clean. Use when the user says "deep review", "/deep-review", or wants a thorough post-implementation review.

Shared skill for all deep-plan, deep-execute, and deep-review agents. Provides two capabilities: (1) Surface past user corrections relevant to the current task, and (2) Save new corrections when the user overrides an agent's output. Feedback is stored in docs/feedback/ as committable markdown files with a searchable index.

Behavioral directive for genuine intellectual engagement. Prevents yes-man behavior — challenges wrong ideas, takes clear positions, names trade-offs honestly. Referenced by other skills and agents that involve discussion, review, or decision-making.

Advisory skill for vision alignment. Maintains docs/product-vision.md as a single source of truth for what the product is and isn't. Advisory — never blocks a pipeline.

PROACTIVE: Enforces zero test/lint/build failures at all times. Auto-triggered after any code modification task completes. No failure is ever dismissed as pre-existing or out-of-scope. Every failure gets fixed, every time.

How to use cmux tools to interact with other terminal surfaces — Claude Code instances, pi agents, browsers. Triggers when you need to find, read, send to, or converse with another terminal surface. Always prefer cmux tools over bash commands like cmux list-pane-surfaces, ps, or cmux tree.

Reference for pi's built-in tools and extensions. Auto-loads when the agent needs to list surfaces, talk to other terminals, manage tasks, use sub-agents, interact with cmux, view files, check isolation, or any operation that has a dedicated tool. Use whenever the agent considers using bash to replicate functionality that already exists as a pi tool or extension.

Discovers and bridges skills from Claude, Cursor, Copilot, and other ecosystems into pi. Use /meta-skills to list and scan external skills.