blindspot

Arguments. [area-or-domain]. e.g., /discovery:blindspot geofencing, /discovery:blindspot payments module, /discovery:blindspot <domain-vocabulary>

Repository context. Gather first

Collect these with individual Bash calls, one command per call, never combined into a single invocation:

• Current branch, git branch --show-current
• Project root, git rev-parse --show-toplevel

Treat a failure (not a repository, git unavailable) as an unknown value and carry on. Keep these as separate body Bash calls rather than pre-compute lines (gather-block convention, "The pre-compute block runs as one shell invocation").

These values orient this session only; resolve files against the project root while working.

Workflow

This skill builds the USER's knowledge, not the agent's: the deliverable is the user's understanding plus one improved implementation prompt, calibrated to their disclosed starting point rather than a fixed depth.

1. Intake. Ask the user's starting point first (one question). What they already know bounds which gaps are worth surfacing.
2. Scan. Two lanes, chosen by what is unfamiliar:
   • Codebase lane. Read the target area (the codebase-reading, git-history, and project-structure dimensions in ${CLAUDE_PLUGIN_ROOT}/skills/explore/reference/workflow.md) looking specifically for things the user's framing missed: existing patterns they'd duplicate, constraints they'd violate, historical decisions they'd re-litigate, adjacent code their change would break.
   • Domain lane. Build a lightweight vocabulary ladder grounded in sources fetched this session (repo files, official docs), never bare training recall. Right after the ladder, add a block of the quality axes experts judge the work by, with a worked good-vs-bad example per axis, each grounded in a source fetched this session. The axes are their own block ahead of the cards, not a card type. Each axis becomes a prompt-fix line in the improved prompt, which the user keeps or deletes like any other fix. A criterion the user's framing never named may also appear as a Missing concept card.
3. Output. Blindspot cards, then ONE improved implementation prompt assembled from their prompt-fix lines (Output format below).
4. Escalate when depth warranted. A domain too deep for a lightweight ladder gets a recommendation to learn it in a structured way: /education:teach topic <domain> when it is among the available skills, otherwise proper external research, /discovery:research invoked via the Skill tool.

Output format

Present each blindspot as a card:

• Type, one of four: Landmine (the change would break something non-obvious), History (a constraint whose reason the code no longer shows), Convention (an unwritten team rule the work must follow), or Missing concept (a domain idea the user's framing never named).
• Gap, the specific thing the user's current framing did not account for.
• Why it matters here, the concrete consequence in this codebase or domain, not a generic caution.
• Prompt-fix, a single copyable line the user can drop into their prompt to close the gap.

Then assemble every prompt-fix into ONE improved implementation prompt, wrapped in clear copy-start / copy-end markers so the exact text to reuse is unambiguous.

End with one scan-scope disclosure line: which lane(s) ran and what was and was not scanned (areas read, sources fetched). One line, not a methodology dump.

For a long card list, offer an interactive view in one sentence (cards filterable by type, a tick for each that was news, a copy-out of the reaction). The page is never written beside the cards, which stay the record. When the reader accepts:

1. Resolve the medium through the shared resolver. Add --argument "<medium>" only when the reader named a medium in the request. <option> is this plugin's medium option, ${user_config.medium}; pass "" when that reads empty or as an unsubstituted placeholder. Read each option value from this text; never put a ${user_config.*} string in a shell command.

   node "${CLAUDE_PLUGIN_ROOT}/lib/view-settings.mjs" resolve --lane discovery:blindspot --option "<option>"
   

   Take the medium from its JSON line only, never from a rendered-views settings file. The step is done when its report line and each notice line are printed as they stand. When node is missing or the resolver exits non-zero, build nothing and say in one line why the view was not built.

2. page false: build nothing. The step is done: the cards stand.

3. page true: build the page only with ${CLAUDE_PLUGIN_ROOT}/scripts/build-view.mjs, never by hand. With publishes true, publish the printed file with the Artifact tool and give the link; when that tool is unavailable, hand back the path and say why in one line. Otherwise hand back the path. The step is done when the reader has the link or the path.

Build command, data shape, build failures and reply ids: ${CLAUDE_PLUGIN_ROOT}/reference/rendered-view.md, steps 1, 2, 4 and 6 and its "Discovery: blindspot" section. The resolver above replaces steps 3 and 5.

This skill makes no changes and runs no open-ended external research: the domain lane fetches official docs only to ground the ladder and quality axes, and anything deeper routes to /discovery:research, invoked via the Skill tool. It does NOT write EXPLORE.md. When the scan's findings also serve as stage-1 codebase exploration, offer to hand off to /discovery:explore, invoked via the Skill tool, to persist the EXPLORE.md artifact.

Next

• The scan doubles as stage-1 codebase exploration: /discovery:explore <area>.
• The prompt is sharpened and ready to scope: /planning:interview.
• Unsure where this leaves the work, or arrived mid-flow: /session-flow:workflow.

Gotchas

• Presenting training recall as domain fact. The domain lane grounds its vocabulary ladder and quality axes, good-vs-bad examples included, in sources fetched this session (repo files, official docs).
• Surfacing the agent's gaps instead of the user's. Cards name what the USER's framing missed, calibrated to their intake disclosure, not a generic audit of the area.
• Generic cautions in "why it matters". Each card's consequence is concrete to this codebase or domain; a caution that would read the same in any repo is not a blindspot.
• Writing an artifact by reflex. No EXPLORE.md unless the user opts into the explore handoff.