discovery

Surface the USER's unknown-unknowns before they work in unfamiliar territory, an unfamiliar codebase area OR an unfamiliar domain vocabulary, and coach a sharper prompt. Scans for gaps the user's framing missed, emits one blindspot card per gap (the gap, why it matters here, a copyable prompt-fix line), then assembles the fixes into one improved implementation prompt. Use when: 'what am I missing', 'find my blindspots', 'what do I not know here', 'sharpen this prompt', 'I am new to this area', or about to work in an unfamiliar area and the goal is a better prompt, not the codebase handoff artifact /discovery:explore produces.

Explore the local codebase before making changes. Read code, trace dependencies, scan git history, discover tests, and audit build and tool configuration, persisting an EXPLORE.md index plus sidecars. Dispatches a fresh-context subagent by default so the file reads stay out of the main conversation. Use when: 'explore the codebase', 'what exists for X', 'how does this work', 'trace the dependencies', 'what tests cover this', or as step 1 before any code change. Skip when: the ask is a bare locate question ('where is X', 'what calls Y') answered by a path and a line number, dispatch the built-in Explore agent instead of this skill. Skip when: the question is why a thing was built the way it was rather than what it is or how it works. Reconstructing the reasoning behind a past decision from review discussion, tickets, and design documents is '/discovery:trace-intent'; this skill's git mode reports what changed and by whom, not the intent behind it.

Multi-source external research in chained phases, corpus enumeration, broad, targeted + falsification, preferred sources, with per-claim source tiers, recency checks, a coverage ledger, and a binary outcome gate before presenting. Dispatches a fresh-context subagent by default so the research transcript stays out of the main conversation, with a documented inline escape hatch. Use when: 'research this', 'verify a technical claim', 'evaluate libraries or approaches', 'compare X vs Y', 'is this still current', 'find the authoritative source', 'what do the official docs say', or grounding any decision in current authoritative sources instead of training data. This is the right skill for a single topic, including a small one. For a multi-topic or workflow-driven pass that fans one question across several dispatches, use research-deep, which layers tiered execution on this same discipline.

Dispatch deep external research to the heaviest isolated execution tier available, a workflow engine, an isolated subagent, or inline as last resort, keeping the main conversation clean while the full research discipline runs. Itself runs in main context, the only place both the Workflow tool (absent from every non-fork subagent) and a dependable Agent spawn are guaranteed. Use when: 'deep research', 'research these N topics', 'broad multi-source research', 'compare these tools thoroughly', 'migration research', 'exhaustive research on X'; for a single small lookup use the research skill directly, which already dispatches its own subagent.

Verify or configure where discovery artifacts land in this repository: report the effective topic-docs concern, or persist it to the tracked .claude/topic-docs.yaml. Use when: 'set up discovery', 'configure the discovery plugin', 'is discovery configured', 'discovery setup', 'where do EXPLORE.md / RESEARCH.md land', or a discovery skill reports missing or thin config. Actions: check (read-only, default) | apply (persist the concern file). Re-runnable. Safe to invoke again.

Reconstruct WHY something was built the way it was, from evidence outside the code, review discussion, tickets, design docs, incident records, and report what could not be found as a first-class result. Grades every claim on an intent-evidence tier (Direct / Supported / Inferred / Speculative / Unknown) and cites each one. Use when: 'why was this built this way', 'why did we pick X over Y', 'what were they thinking', 'design rationale', 'what problem was this solving', 'why does this still exist', 'archaeology on this decision'. Skip when: the question is what the code DOES or where it lives (that is '/discovery:explore', whose git mode owns repo-local history); when the question is whether a convention should STILL hold rather than why it was adopted (that is '/discipline:reason-dont-recite'); or when the answer is a current external fact rather than a past decision (that is '/discovery:research').