agent-skills

Compact the current conversation into a handoff document for another agent to pick up.

Build a non-ambiguous, complete, provenance-tagged product spec for a feature area — the behavioral oracle that prove-behavior tests against. Researches every reachable source (code, live product, sibling specs, Jira, Confluence, Slack, git history) to build a mental map, then interrogates the user exhaustively to resolve EVERY ambiguity before writing. Use when: (1) user says 'feature-spec', 'write the spec for X', 'build the product spec', 'spec out this feature', (2) before running prove-behavior on a feature that has no spec yet, (3) onboarding onto a feature and wanting the ground-truth contract. One spec per feature AREA (e.g. all of the settings screen, or checkout, or the text editor). Output: docs/features/<area>/product-spec.md.

Append a verbatim entry to a personal, per-repo, gitignored devlog (DEVLOG.md), and seed talking-point nudges from recent git/PR activity and unresolved threads in the previous entry. Use when the user says "create devlog", "devlog", "add a devlog", "dev log entry", "log this", "/create-devlog", or wants to record what they are working on / rant about progress. The user usually dictates via speech-to-text, so the log is recorded WORD-FOR-WORD with only obvious typo and speech-to-text fixes, never summarized, reworded, or analyzed.

Append a tech debt entry to a project tech-debt log (docs/tech-debt-log.md) based on the current conversation. Use when: (1) the user says 'add tech debt', 'log tech debt', 'log this as tech debt', 'note this for the tech debt log', '/add-tech-debt', (2) the user wants to record an architectural smell, coding-practice issue, or non-blocking design flaw discovered while working on something else, (3) the user describes a self-healing or workaround-able issue that signals a deeper design mistake worth logging. Triggers on tech debt logging, architectural smell capture, code quality issue recording.

Systematic pre-implementation analysis that finds unknown unknowns — usability gaps, broken flows, state conflicts, concurrency hazards, error dead-ends, and edge cases BEFORE code is written. Fire-and-forget: runs all phases autonomously and delivers a severity-ordered report. Works across stacks — frontend, backend, CLI, mobile, APIs. Use this skill when: (1) before implementing any feature that changes existing behavior, (2) before adding async/background processing to what was synchronous, (3) when a feature touches shared state (global stores, caches, databases, queues), (4) when a feature has multiple entry points or triggers, (5) when moving functionality between surfaces or layers, (6) when the user asks to 'think through' or 'analyze' a feature before building it, (7) when reviewing a plan for hidden failure modes. Triggers on: 'analyze this feature', 'think through the flow', 'what could go wrong', 'check for edge cases', 'will this break anything', 'find problems with this', 'unknown unknowns', flow analysis, impact analysis, pre-implementation review.

Adaptive thinking partner that helps clarify, challenge, and refine ideas through persistent questioning. Auto-detects domain (product, architecture, debugging, process, general) and user mode (exploring, deciding, refining) to adapt question style. Actively pushes back on weak reasoning — flags contradictions, challenges assumptions, stress-tests claims. Produces context-appropriate artifacts when done (design doc, hypothesis list, decision matrix, or key insights). Use this skill when: (1) brainstorming or exploring an idea before implementation, (2) requirements are vague and need clarification, (3) making architectural or product decisions, (4) debugging and need to form hypotheses, (5) refining an approach that's mostly decided. Triggers on: 'brainstorm', 'clarify', 'think through', 'explore', 'help me figure out', 'what should I consider', 'let's think about', 'what could go wrong', 'help me decide'.