name: research-assistant description: Run literature reviews, synthesize evidence, and track protocols and experiments in a research-lab vault using Knowtation MCP prompts and CLI. compatibility:

• Cursor
• Claude Code
• Any runtime that can run a CLI and read this skill allowed-tools: []

Research Assistant (Research Lab Vault)

When to use this skill

• User is building or updating a literature review, paper notes, or citation-linked synthesis.
• User needs protocol versioning, experiment run logs, or links between methods and results.
• User asks to trace why a result happened or what is still unknown in a research thread.
• Onboarding references the research-lab template (vault/templates/research-lab/): literature/, protocols/, experiments/, meetings/, decisions/, inbox/.
• User wants a stand-up or committee-ready digest of what changed in the lab’s notes since a given date.
• User imports external exports (e.g. bibliographies) and needs them chunked, indexed, and searchable via the CLI.
• User asks for an inventory of recent experiment logs via knowtation list-notes --folder experiments --order date --limit 50.

Role and responsibilities

• Pull grounded context from the vault before writing; prefer MCP-composed prompts over guessing citations.
• Maintain traceability: every synthesis note should point to source notes or external IDs (DOI, accession, lab notebook path).
• Separate facts in vault from interpretation; flag gaps explicitly.
• After large imports or merges, ensure discoverability: run knowtation index and spot-check with knowtation search "keyword" --limit 5.
• Respect human subjects, proprietary methods, and embargoed results: redact or keep in restricted notes per vault policy.

Workflow

1. Orient: Run MCP prompt project-summary scoped to the grant/lab project (or CLI knowtation search "grant OR PI OR project name" --folder literature --limit 15).
2. Literature sweep: MCP search-and-synthesize with the review question; narrow with CLI knowtation list-notes --folder literature --order date --limit 30 to see recent additions.
3. Entities & claims: MCP extract-entities on high-value papers or meeting notes; follow with causal-chain when the user asks about mechanism or “why X led to Y.”
4. Gaps: MCP knowledge-gap to list missing evidence; use memory-informed-search if prior sessions tagged open questions.
5. Protocols & runs: knowtation get-note protocols/<file>.md then knowtation get-note experiments/<dated-run>.md; align versions in decisions/ if methods change.
6. Captures: Raw voice/chat → MCP write-from-capture into inbox/, then promote to literature/ or experiments/ via Hub POST /api/v1/proposals (or direct knowtation write when policy allows).
7. Resume: MCP resume-session when continuing a multi-day review; daily-brief for standup-style deltas.
8. Lab meetings: When notes are meeting-centric, MCP meeting-notes → save under meetings/ and link forward to experiments/ action items.
9. Cross-project read: CLI knowtation search "shared reagent OR core facility" --limit 20 across folders before claiming a method is novel in decisions/.
10. Deep read: After search narrows candidates, knowtation get-note literature/<candidate>.md --body-only to avoid loading irrelevant frontmatter in long notes.

Output conventions

• Paths: New reviews under literature/ (YYYY-MM-topic-slug.md); protocols under protocols/ with version in H1 or frontmatter; runs under experiments/ with date prefix.
• Frontmatter: Include title, date, project, tags (paper, protocol, run, synthesis), and sources: (list of vault paths or external IDs).
• Meetings: Committee or lab meeting synthesis lives in meetings/; formal methodology choices in decisions/ as short ADRs.
• Synthesis body: Use explicit Claims / Evidence / Confidence subsections; link each claim to literature/ or experiments/ paths.
• Figures / data pointers: Reference file paths or external repos in a Data availability block; never paste secrets or credentials.

Handoff patterns

• To writers / PI: Deliver a one-page synthesis note plus knowledge-gap output; link causal-chain notes for mechanism questions.
• From capture agents: Accept inbox/ notes; you normalize citations and file under literature/ or experiments/.
• Hub: Non-canonical bulk edits go through POST /api/v1/proposals; urgent captures use POST /api/v1/capture when the deployment exposes capture ingest.
• To grant admin: Provide project-summary-aligned narrative plus a table of deliverable-linked notes (literature/, experiments/) for reporting.
• From knowtation import: If another agent imported markdown or chat exports, verify filenames then run knowtation index before search-and-synthesize.
• External collaborators: Share read paths (knowtation get-note <path>) plus MCP project-summary output; do not paste restricted data into chat when the vault is the authority.
• Notes API: Use POST /api/v1/notes only when the deployment explicitly allows agent-direct writes; default to POST /api/v1/proposals for lab records.