engineering-principles
Apply named principles to coding or review decisions, such as fix root causes or explain the number. Read only relevant files; use AGENTS.md for repository rules, prove-it-works or vet-measurement for verification procedures.
38,472 Skills
Search uses this published catalog; rankings and counts are snapshots.
Apply named principles to coding or review decisions, such as fix root causes or explain the number. Read only relevant files; use AGENTS.md for repository rules, prove-it-works or vet-measurement for verification procedures.
Explain current code with file references and explorer subagents. Use for unfamiliar code or how-it-works questions; use explain-why for history or rationale, or a bare explorer Task for quick lookups.
Reproduce a defect, fix its root cause, and verify the repro with prove-it-works. Use for defect reports; use Normal mode for direct fixes needing no repro-first loop, or tune-perf for performance regressions.
Recap changes, files, decisions, and pending user calls since the last message or across past sessions. Use when returning to work; use droid search, droid resume, or session-navigation to find, list, or resume sessions.
Deepen a vague or broad prompt without expanding scope before Spec mode or a Mission. Use Spec mode directly once the request is sharp.
Verify each PR in an isolated worktree, run simplify, and open it with gh. Use for completed work ready for review; use bare gh pr create if those passes are unnecessary, or watch-pr for an open PR.
Cut filler, hedging, and stock phrasing while applying technical-writing rules. Use for READMEs, docs, PRs, commits, and comments; use simplify for code reuse, quality, and efficiency.
Check an installed AutoBot workspace with its read-only doctor and focused runtime diagnostics. Use for installation health, capability readiness or a requested health report; a check alone does not authorize repairs.
Coordinate authorized parallel workers on bounded AutoBot subtasks and verify their outputs. Use when the user requests delegation or the current environment explicitly permits it; keeps responsibility with the lead.
Keep a substantial AutoBot assignment recoverable with one durable objective, exact checkpoints, supported continuation and verified deliverables. Use when the user asks for persistent follow-through or to finish work across sessions.
Retrieve and update scoped AutoBot memory and apply user corrections with provenance. Use when the user asks to remember a durable fact, correct a saved preference or carry a lesson into later work.
Stagnation detector + persona switcher. Triggers when the same file is edited 3+ times for the same issue, the same hypothesis fails 3+ times in mcp_think, the auditor returns BLOCK twice in a row, or test pass rate plateaus across attempts. Detects 4 stagnation patterns (Spinnin
Design cooking — the co-creation front-end that DISCOVERS a design instead of assuming one exists. USER-INVOKED ONLY — triggers when the user types /cook, /cook <surface>, says "cook the design", "let's cook this", "help me figure out the look", "i don't know what this should loo
Relentless one-question-at-a-time interrogation that drives every unresolved decision out of a plan, design, or fuzzy idea BEFORE prompt-transform reformats it and any code or formal plan is written. USER-INVOKED ONLY — triggers when the user types /grill, /grill <topic>, says "g
Drive any Windows GUI application on the owner's real PC and PROVE what was done. A "fake human" that opens apps, reads their UI Automation tree, clicks/types/scrolls through them, and journals every action frame-bound and replayable — the desktop generalization of the playtester
Interactive grounded Q&A. Triggers when user types /oracle [question], /ask [question], or asks any deep factual question requiring web grounding (current API behavior, library comparison, framework patterns, App Store rules, recent ecosystem changes, etc). Spawns the researcher
Catalog captured:2026-11-21