From dotclaude
Plan a feature through codebase discovery and iterative dialogue before writing any code. Runs a scout agent to map the codebase, works through decisions one at a time, then compiles a decision spec. Use AFTER product-reasoning (should we build it?) and BEFORE code-mode (building it).
How this skill is triggered — by the user, by Claude, or both
Slash command
/dotclaude:feature-discoveryThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Read the shared philosophy: `${CLAUDE_PLUGIN_ROOT}/skills/shared/philosophy.md`
Read the shared philosophy: ${CLAUDE_PLUGIN_ROOT}/skills/shared/philosophy.md
You are running feature discovery for:
$ARGUMENTS
ultrathink
Use the Agent tool to run the feature-scout subagent.
Load its instructions from: ${CLAUDE_PLUGIN_ROOT}/agents/feature-scout.md
Pass the feature description above as the agent's task.
Wait for the full report before continuing. Store it — you will reference it throughout.
Present the discovery findings to the user. Keep the summary tight:
Don't dump the full report. Highlight what matters for decision-making. If the user asks to see the full report, show it.
Then work through decisions iteratively:
High confidence — obvious from codebase patterns, or simple with clear precedent: Suggest and ask for confirmation. One sentence on why. "The codebase validates at the handler level using X pattern. I'd do the same here. Agree?"
Low confidence — ambiguous, trade-offs, or no clear precedent: Present 2–3 options with concrete trade-offs. Ask the user to decide. "Should this be synchronous or async? Sync is simpler but blocks the response. Async would need a queue we don't have yet."
Impact on existing behaviour: Flag explicitly before proceeding. "This would change how Y works. Is that intended?"
Rules:
Once all decisions are resolved (or the user says they're done), compile:
What it does Plain description from the user's intent and confirmed decisions.
Components affected From the discovery report. Each entry: component + what changes.
Decisions
| Decision | What was decided | Source |
|---|---|---|
| [topic] | [decision] | suggested+confirmed / explicitly stated |
Simplicity check Is the proposed approach the fewest moving parts that solves the problem? Flag anything not earning its complexity.
Open items Unresolved questions, if any. If none, say so.
Acceptance criteria Derived from confirmed decisions. Each criterion must be falsifiable.
Every line in this spec traces to something the user stated or confirmed. Nothing invented. This spec is the input to test-writing and code-mode — not a plan, a decision record.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub realdragonium/dotclaude --plugin dotclaude