菜单
Company research using Exa search. Finds company info, competitors, news, financials, LinkedIn profiles, builds company lists. Use when researching companies, doing competitor analysis, market research, or building company lists. NOT for individual-person lookups (use /people-research).
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Interactive first-run setup for a fresh fork of the second-brain-os — THE entry point every fork user runs once to configure their identity, name their assistant's persona (no default name — fork users always name their own), pick a design system, create the PARA workspace, and write the Configuration token block in root CLAUDE.md. A short, plain-language guided walkthrough for knowledge workers (~8-10 min) with AskUserQuestion gates and a brief why-it-matters intro on each step; nothing is installed, nothing is committed, every write asks first. When core setup finishes, it offers two OPTIONAL go-further steps — learning your writing voice from your own sent Slack/Gmail messages (with consent), and bringing your existing work in via `/migrate-work`. Detects re-runs via `setup_completed:` and refuses gracefully. Use after cloning the repo, or to reconfigure identity / persona / workspace — phrases like "/bootstrap", "I just cloned this", "first-time setup", "configure the assistant". Tiger invariants T1-T4 (n
Bring a user's EXISTING work into their second-brain-os PARA workspace by reading their past Claude Cowork sessions (local transcripts on disk) — discovering the projects they've actually been working on, then scaffolding the chosen ones as organized PARA entries (1-Projects or 4-Resources) with a CLAUDE.md + memory.md seeded from the session context. Read-only discovery + a fan-out of scout subagents (one per discovered project) + a SINGLE approval gate, then copy-never-move writes. Offered at the end of `/bootstrap` and re-invokable anytime. Use when the user says "/migrate-work", "bring in my existing work", "import my projects", "I have existing projects to bring over", "organize my past work", "migrate my cowork sessions". Hard invariants live in SKILL.md body — copy-never-move, read-only discovery, one approval gate, no fabrication, runnable code routes to the coding bucket (`<workspace.coding>`), never auto-commit.
OS reference manual — for fork users in their first 30 days AND for Claude Code self-correcting mid-session on an OS-shaped mistake. Routes the question to its canonical source file, reads at runtime, answers with file:line citations. Topics — PARA, Configuration tokens, memory model, contacts schema, tools, lifecycle, design system, locked decisions, capability index, trigger-phrase map, Operating Principles. **Claude SHOULD consult this skill** before answering questions involving OS configuration values, workspace paths, or canonical skill behavior — especially before writing file paths into responses or scaffolding new files. Read-only by default; `/os-guide --sync` is the sole mutation mode and is gated behind explicit user approval. PRECEDENCE — `/bootstrap` for first-run setup; `/find` for topic search across user content; `/contact` for people; `/os-guide` for the OS itself.
Moves a completed project from `<workspace.root>/<workspace.projects>/<slug>/` to `<workspace.root>/<workspace.archive>/<slug>/` and flips its frontmatter `status: active` → `status: done`. All paths come from the Configuration section in root CLAUDE.md — read those first. Use this whenever the user says they're done with a project, wants to wrap it up, or wants to clean up their active list — phrases like "archive the X project", "I'm done with X", "wrap up X", "move X to archive", "X is finished — clear it out", "shelve X for now". Optional one-paragraph retro gets appended to `memory.md` before the move so the decision context survives. Trigger even when the user doesn't say "archive" explicitly — finishing/wrapping/clearing language for a known project should invoke this skill rather than letting them `mv` by hand (which would skip the status flip and break `/prune-projects` later).
Upload files (screenshots, diagrams, PDFs, exported designs) as attachments to a Jira issue or a Confluence page. Works for either target — pick the subcommand. Uses Atlassian REST API directly via Python urllib + multipart — the official Atlassian MCP has no attachment-upload tools, so this skill exists to fill that hole. Auth via ATLASSIAN_BASIC_AUTH env var (classic API token only — scoped tokens cannot upload). Use whenever the user wants to add an image, screenshot, design export, or any binary file to an existing Jira ticket or Confluence page — phrases like "attach this screenshot to PROJ-450", "upload my-diagram.png to the design page", "add this PDF to the ticket", "embed this image in Confluence". Pairs naturally with `/confluence-publish-markdown` (which embeds external URLs only — no native upload) and with `/scaffold-engineering-project` (when slice descriptions need supporting images).
Deep, iterative, hypothesis-driven research that loops until confident — not a single-pass lookup. Builds on /research by spawning multiple hypothesis workers per iteration, accumulating evidence in Ouros sessions, looping until confident or the user stops. Use when the user says 'research this deeply across multiple passes', 'run iterative research on', 'hypothesis-driven investigation of', 'loop until you understand X', 'deep-dive with synthesis'. NOT for single-session exploration (use /research) or implementation (use /autonomous).
| name | company-research |
| description | Company research using Exa search. Finds company info, competitors, news, financials, LinkedIn profiles, builds company lists. Use when researching companies, doing competitor analysis, market research, or building company lists. NOT for individual-person lookups (use /people-research). |
| allowed-tools | Task mcp__exa__web_search_advanced_exa |
ONLY use web_search_advanced_exa. Do NOT use web_search_exa or any other Exa tools.
Never run Exa searches in main context. Always spawn Task agents:
No hardcoded numResults. Tune to user intent:
Exa returns different results for different phrasings. For coverage:
Use appropriate Exa category depending on what you need:
company → homepages, rich metadata (headcount, location, funding, revenue)news → press coverage, announcementspeople → LinkedIn profiles (public data)type: "auto") → general web results, deep dives, broader contextStart with category: "company" for discovery, then use other categories or no category for deeper research.
When using category: "company", these parameters cause 400 errors:
includeDomains / excludeDomainsstartPublishedDate / endPublishedDatestartCrawlDate / endCrawlDateWhen searching without a category (or with news), domain and date filters work fine.
Universal restriction: includeText and excludeText only support single-item arrays. Multi-item arrays cause 400 errors across all categories.
Public LinkedIn via Exa: category: "people", no other filters.
Auth-required LinkedIn → use Claude in Chrome browser fallback.
Auto-fallback to Claude in Chrome when:
web_search_advanced_exa {
"query": "AI infrastructure startups San Francisco",
"category": "company",
"numResults": 20,
"type": "auto"
}
web_search_advanced_exa {
"query": "Anthropic funding rounds valuation 2024",
"type": "deep",
"numResults": 10,
"includeDomains": ["techcrunch.com", "crunchbase.com", "bloomberg.com"]
}
web_search_advanced_exa {
"query": "Anthropic AI safety",
"category": "news",
"numResults": 15,
"startPublishedDate": "2024-01-01"
}
web_search_advanced_exa {
"query": "VP Engineering AI infrastructure",
"category": "people",
"numResults": 20
}
Return: