deep-researchlisted

# Deep Research Answer questions from real, read sources — not memory. Every claim traces to a URL. `S=~/.claude/skills/deep-research/scripts/fc.py` (run `python3 $S <cmd>`). The key is read automatically from `.claude/.env.local`, `~/.claude/.env.local`, or `$FIRECRAWL_API_KEY`. ## Workflow **1. Scope.** Restate the question in one line. Pick a depth (default: standard). Ask a clarifying question *only* if two readings would produce genuinely different research — otherwise state your assumption and go. | depth | queries | read | ~credits | |-------|---------|------|----------| | quick | 3 | 4–5 | ~15 | | standard | 6 | 10–12 | ~35 | | deep | 8–12, two rounds | 18–25 | ~70 | **2. Plan.** Write 3–12 sub-queries as a visible plan before running anything. Cover different *angles*, not rephrasings: definition/primary source, current state, numbers and benchmarks, criticism and failure modes, comparisons, recency. Use the words a source would use, not the words the user used. **3. Search** — one call, all queries run in parallel: ```bash python3 $S search "sub-query one" "sub-query two" "sub-query three" \ --limit 6 --topic "short label" [--recent month] [--news] ``` Prints a numbered source table and the RUN directory. Keep that path — every later command needs `--run`. **4. Select.** From the table, pick sources for *diversity and authority*: primary docs, official announcements, papers, and specs beat listicles and SEO blogspam. Include at least one skeptical or com