research-protocol
Gather facts from the web using evidence-first approach with mandatory citations. Use when researching topics, answering factual questions, or any task requiring web-sourced evidence.
What this skill does
## Available research tools - **web-search(query, limit)** — DuckDuckGo search. Returns titles, URLs, and descriptions. - **web-fetch(source)** — Fetch a URL or local file and convert to Markdown text. Use on URLs found via web-search. - **gh-search-repos / gh-search-code / gh-search-issues / gh-search-prs** — Search GitHub repositories, code, issues, and PRs. - **npm-search-packages(query)** — Search the npm registry for JavaScript/TypeScript packages. - **pypi-search-packages(query)** — Search PyPI for Python packages. - **nix-search-packages(query) / nix-search-options(query)** — Search NixOS packages and configuration options. - **hm-search-options(query)** — Search Home Manager configuration options. - **hf-search-models(query)** — Search Hugging Face models. ## Research Protocol (evidence-first) 1. Decompose the question into one or two unknowns. Write them down in your first reply. 2. For each unknown, search with the appropriate tool above, then fetch relevant pages with web-fetch, then call evidence-add (one fact per entry). 3. Do NOT state a fact that isn't backed by an evidence-add id. 4. Before answering, call evidence-list. Your answer must reference at least one id. 5. If evidence-list is empty, you are not ready to answer — go back to step 2. ## Rules - NEVER state a fact that isn't backed by an evidence-add id - ALWAYS call evidence-list before answering - Your answer must reference at least one evidence id - If evidence-list is empty, go back to gathering evidence - After 3+ search refinements with no usable evidence, say "insufficient evidence" instead of guessing ## Stop conditions - You have an evidence id for every claim in your answer → ANSWER - You have tried 3+ search refinements with no usable evidence → say "insufficient evidence" ## Example workflow ``` web-search(query="Turing machine invention date") → returns results including Wikipedia URL web-fetch(source="https://en.wikipedia.org/wiki/Turing_machine") evidence-add(source="...", note="Turing machine introduced in 1936", snippet="...") evidence-list() → Answer with citations: e1 ``` ## Workflow 1. **Decompose**: Break the question into 1-2 unknowns 2. **Search**: Use web-search, gh-search-\*, npm-search-packages, pypi-search-packages, or other domain tool 3. **Fetch**: Use web-fetch on promising URLs to get full page content 4. **Save evidence**: Call evidence-add for each fact found (one fact per entry) 5. **Verify**: Call evidence-list before answering; answer must reference at least one ID 6. **Stop condition**: Answer when all claims have evidence; say "insufficient evidence" after 3+ failed searches
Related in Productivity
gitea-workflow
IncludedOrchestrate agile development workflows for Gitea repositories using the tea CLI. Use when working with Gitea-hosted repos and asking to 'run the workflow', 'continue working', 'what's next', 'complete the task cycle', 'start my day', 'end the sprint', 'implement the next task', or wanting guided step-by-step development assistance. Keywords: workflow, orchestrate, agile, task cycle, sprint, daily, implement, review, PR, standup, retrospective, gitea, tea.
microsoft-graph-gateway
IncludedRoute Microsoft Graph work in this workspace. Use when users want to read or write Outlook mail, calendar events, contacts, OneDrive or SharePoint files, Teams, Planner, To Do, users, groups, directory data, or arbitrary Microsoft Graph endpoints from VS Code. Prefer WorkIQ for common read scenarios. Use Microsoft Graph for write actions and gap-read scenarios that need exact Graph properties, filters, permissions, or endpoints.
copilotkit
IncludedUse when building with CopilotKit — setup, development, integrations, debugging, upgrading, or contributing. Routes to the appropriate specialized skill based on the task.
wordly-wisdom
IncludedProvides calibrated decision analysis using Charlie Munger-style multiple mental models, inversion, incentive mapping, circle-of-competence checks, misjudgment audits, second-order effects, and forecast updates. Use when the user asks for an oracle take, a hard call, a decision memo, a premortem, an outside view, a red-team, a sanity-check, what am I missing, think this through, or wants a strategy, hire, investment, plan, product, partnership, or major life choice analysed. Avoid for simple factual lookups or time-sensitive legal, medical, or market questions without fresh evidence.
swain-session
IncludedSession management and project status dashboard. Owns the full session lifecycle (start/work/close/resume), focus lane, bookmarks, worktree detection, and tab naming. Also serves as the project status dashboard — shows active epics, progress, actionable next steps, blocked items, tasks, GitHub issues, and recommendations. Worktree creation is deferred to swain-do task dispatch (SPEC-195). Triggers on: 'session', 'status', 'what's next', 'dashboard', 'overview', 'where are we', 'what should I work on', 'show me priorities', 'bookmark', 'focus on', 'session info'.
gandi
IncludedComprehensive Gandi domain registrar integration for domain and DNS management. Register and manage domains, create/update/delete DNS records (A, AAAA, CNAME, MX, TXT, SRV, and more), configure email forwarding and aliases, check SSL certificate status, create DNS snapshots for safe rollback, bulk update zone files, and monitor domain expiration. Supports multi-domain management, zone file import/export, and automated DNS backups. Includes both read-only and destructive operations with safety controls.