surveying-blind-spotslisted

# Surveying Blind Spots The user cannot ask about what they cannot name. Explore the target on their behalf, surface the unknown unknowns, and leave them able to prompt and plan with eyes open. This is a briefing, not a tutoring loop: do not quiz, do not gate on mastery, do not lecture the whole field. ## When to invoke - "Do a blind spot pass on X." "I'm new to this module, what don't I know?" "What are my unknown unknowns about Y?" "Help me prompt better about Z." ## When NOT to invoke - The user wants to be taught and quizzed to mastery. Use `learning:quizzing-a-topic`. - The user is ready to converge a design through Q and A. Use `workbench:brainstorming`; this skill feeds it. - The user wants a cited multi-source report. Use `research:research`. ## Scope the pass - Ask at most one clarifying question, and only if the target is unbounded. Starting fast beats interrogating. - The user often names where context lives (a module path, a branch, a doc folder, an exported thread). Follow those pointers before anything else. ## Codebase pass When the target maps to a repo: 1. Dispatch a cost-efficient subagent (Claude Code: `Explore` agent type; Codex: read-only research agent equivalent; inline if dispatch is unavailable) over the named area: entry points, config surface, and the two or three files everything else imports. 2. Mine the history for pain: `git log --oneline -30 -- <area>`, then read commits whose messages contain fix, revert, workaround, or hack. Repeate