SKILL.md
file-level
1
files
1
commits
0
hotspots
0
🧊 dead
0
💥 blast risk
| 1 | --- |
| 2 | name: research-assistant |
| 3 | description: Run literature reviews, synthesize evidence, and track protocols and experiments in a research-lab vault using Knowtation MCP prompts and CLI. |
| 4 | compatibility: |
| 5 | - Cursor |
| 6 | - Claude Code |
| 7 | - Any runtime that can run a CLI and read this skill |
| 8 | allowed-tools: [] |
| 9 | --- |
| 10 | |
| 11 | # Research Assistant (Research Lab Vault) |
| 12 | |
| 13 | ## When to use this skill |
| 14 | |
| 15 | - User is building or updating a **literature review**, paper notes, or citation-linked synthesis. |
| 16 | - User needs **protocol versioning**, experiment run logs, or links between methods and results. |
| 17 | - User asks to trace **why** a result happened or **what is still unknown** in a research thread. |
| 18 | - Onboarding references the **`research-lab`** template (`vault/templates/research-lab/`): `literature/`, `protocols/`, `experiments/`, `meetings/`, `decisions/`, `inbox/`. |
| 19 | - User wants a **stand-up or committee-ready** digest of what changed in the lab’s notes since a given date. |
| 20 | - User imports external exports (e.g. bibliographies) and needs them **chunked, indexed, and searchable** via the CLI. |
| 21 | - User asks for an inventory of **recent experiment logs** via `knowtation list-notes --folder experiments --order date --limit 50`. |
| 22 | |
| 23 | ## Role and responsibilities |
| 24 | |
| 25 | - Pull grounded context from the vault before writing; prefer MCP-composed prompts over guessing citations. |
| 26 | - Maintain traceability: every synthesis note should point to source notes or external IDs (DOI, accession, lab notebook path). |
| 27 | - Separate **facts in vault** from **interpretation**; flag gaps explicitly. |
| 28 | - After large imports or merges, ensure discoverability: run `knowtation index` and spot-check with `knowtation search "keyword" --limit 5`. |
| 29 | - Respect human subjects, proprietary methods, and embargoed results: redact or keep in restricted notes per vault policy. |
| 30 | |
| 31 | ## Workflow |
| 32 | |
| 33 | 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`). |
| 34 | 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. |
| 35 | 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.” |
| 36 | 4. **Gaps:** MCP **`knowledge-gap`** to list missing evidence; use **`memory-informed-search`** if prior sessions tagged open questions. |
| 37 | 5. **Protocols & runs:** `knowtation get-note protocols/<file>.md` then `knowtation get-note experiments/<dated-run>.md`; align versions in `decisions/` if methods change. |
| 38 | 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). |
| 39 | 7. **Resume:** MCP **`resume-session`** when continuing a multi-day review; **`daily-brief`** for standup-style deltas. |
| 40 | 8. **Lab meetings:** When notes are meeting-centric, MCP **`meeting-notes`** → save under `meetings/` and link forward to `experiments/` action items. |
| 41 | 9. **Cross-project read:** CLI `knowtation search "shared reagent OR core facility" --limit 20` across folders before claiming a method is novel in `decisions/`. |
| 42 | 10. **Deep read:** After search narrows candidates, `knowtation get-note literature/<candidate>.md --body-only` to avoid loading irrelevant frontmatter in long notes. |
| 43 | |
| 44 | ## Output conventions |
| 45 | |
| 46 | - **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. |
| 47 | - **Frontmatter:** Include `title`, `date`, `project`, `tags` (`paper`, `protocol`, `run`, `synthesis`), and `sources:` (list of vault paths or external IDs). |
| 48 | - **Meetings:** Committee or lab meeting synthesis lives in `meetings/`; formal methodology choices in `decisions/` as short ADRs. |
| 49 | - **Synthesis body:** Use explicit **Claims / Evidence / Confidence** subsections; link each claim to `literature/` or `experiments/` paths. |
| 50 | - **Figures / data pointers:** Reference file paths or external repos in a **Data availability** block; never paste secrets or credentials. |
| 51 | |
| 52 | ## Handoff patterns |
| 53 | |
| 54 | - **To writers / PI:** Deliver a one-page **synthesis** note plus `knowledge-gap` output; link `causal-chain` notes for mechanism questions. |
| 55 | - **From capture agents:** Accept `inbox/` notes; you normalize citations and file under `literature/` or `experiments/`. |
| 56 | - **Hub:** Non-canonical bulk edits go through **`POST /api/v1/proposals`**; urgent captures use **`POST /api/v1/capture`** when the deployment exposes capture ingest. |
| 57 | - **To grant admin:** Provide **`project-summary`**-aligned narrative plus a table of **deliverable-linked** notes (`literature/`, `experiments/`) for reporting. |
| 58 | - **From `knowtation import`:** If another agent imported markdown or chat exports, verify filenames then run `knowtation index` before **`search-and-synthesize`**. |
| 59 | - **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. |
| 60 | - **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. |