Free SKILL.md scraped from GitHub. Clone the repo or copy the file directly into your Claude Code skills directory.
npx versuz@latest install 1tangyen-voxelnoir-dissectorgit clone https://github.com/1tangyen/voxelnoir-dissector.gitcp voxelnoir-dissector/SKILL.MD ~/.claude/skills/1tangyen-voxelnoir-dissector/SKILL.md--- name: company-intel metadata: version: 4.0.0 description: Deep-dive research on technology-driven B2B companies. Produces a sourced 8-chapter analysis report + interactive HTML. Covers enterprise SaaS, infrastructure/DevOps, cybersecurity, developer tools, applied AI, industrial tech, and pharma/biotech. Use when user names a company and wants technical due diligence, competitive analysis, or moat validation. --- # company-intel ## Trigger User names a company → start this workflow. ## Core Principle: No Black Boxes Every fact in the output must trace back to an original source the reader can verify. The report is transparent glass, not a black box. If a claim has no source, it doesn't go in. "Undisclosed" is acceptable; unsourced assertions are not. ## Scope Enterprise SaaS · Infrastructure/DevOps · Cybersecurity · Developer Tools · Applied AI/ML · Industrial/Supply Chain Tech · Pharma/Biotech **Not covered**: pure fintech (banking/insurance), climate/energy, consumer internet. ## Workflow ### 1. Identify company and sector Confirm sector with user. Load the relevant sector overlay from `references/analysis-guide.md § IV`. ### 2. Deduplication check Check output path for existing `[slug].en.md` + `[slug].html`. If both exist, skip unless user requests re-analysis. ### 3. Research (see `references/data-sources.md`) **Hard rules**: - Every search query includes the current year - Every data point records its source URL during research - Stop only when the minimum data threshold is met (see data-sources.md § Sufficiency) - On failure (403, paywall, timeout): fall back, don't retry the same class of source ### 4. Write report `[slug].en.md` Structure and standards in `references/analysis-guide.md`. Key constraints: - Clean Markdown. No YAML frontmatter, no H1, no HTML comments. - Opens with a blockquote judgment (2-4 sentences). - Eight chapters: Moat → Product → Business Model → Regulatory → Partnerships → Competition → Team → Judgment. - Every number or strategic fact gets `[Source: [text](url)]` on the next line. - **Chapter VII must list ALL identifiable C-level executives** (CEO, CSO, CTO, CFO, CMO, CBO, CAO, etc.) with name, prior company/role, and domain expertise. No "undisclosed" for roles that exist. If a role doesn't exist, state explicitly (e.g., "No dedicated CFO"). - Run AI-taste cleanup per `references/ai-taste-guide.md` before saving. ### 5. Generate HTML (default: 2 files only) ```bash python3 scripts/start_preview.py <slug> <output_base> python3 scripts/start_deck.py <slug> <output_base> --density=exec ``` Default output is **EN preview + EN exec deck** (2 files). Additional outputs only on request: - `--lang=zh` → Chinese version of each - `--density=full` → Full deck (20-30 slides) ### 6. Deliver Report file paths + `file://` preview link to user. ## Output Files **Default (always generated):** | File | Content | |------|---------| | `[slug].en.md` | English report | | `[slug].html` | Interactive HTML report | | `[slug].deck-exec.html` | Executive deck (8-12 slides) | **On request only:** | File | When | |------|------| | `[slug].zh.md` | `--lang=zh` | | `[slug].zh.html` | `--lang=zh` | | `[slug].deck-full.html` | `--density=full` | | `[slug].zh.deck-exec.html` | `--lang=zh` | | `[slug].zh.deck-full.html` | `--lang=zh --density=full` | ## Output Path Resolution 1. User-specified path → `BusinessModelCraft/[slug]/` 2. `$BUSINESSMODELCRAFT_OUTPUT_BASE` or `$BUSINESSMODELCRAFT_OUTPUT_DIR` 3. Default: `~/Documents/BusinessModelCraft/[slug]/` ## Reference Files | File | Contains | |------|----------| | `references/analysis-guide.md` | Eight-chapter framework, sector overlays, writing standards, opening/title guidance | | `references/data-sources.md` | Tiered search strategy, access patterns, error handling, sufficiency thresholds | | `references/ai-taste-guide.md` | Post-generation cleanup rules (banned words, structure patterns, em-dash limits) | ## Pre-submission Checklist - [ ] Every number has a `[Source: ...]` line - [ ] No fabricated data — gaps marked "undisclosed" - [ ] Pipeline/product table covers all known items - [ ] AI-taste scan completed (see `references/ai-taste-guide.md`) - [ ] HTML generated and path reported to user