Use when running a formal research project on typed research records — creating a project, importing sources, capturing evidence with locators, building supported claims, auditing provenance, and generating an evidence-backed outline via the research_* tools.
Scanned 9/4/2026
Install to Claude Code
npx -y skills add cavi-ai/obsidian-agent --skill research-workbench --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Research Workbench?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/cavi-ai-research-workbench)More formats (shields.io, HTML) on the badges page.
---
name: research-workbench
description: Use when running a formal research project on typed research records — creating a project, importing sources, capturing evidence with locators, building supported claims, auditing provenance, and generating an evidence-backed outline via the research_* tools.
portable: false
---
# Research workbench
> **Claude adapter only.** The typed `research_*` record API has no equivalent
> in the official Obsidian CLI. Keep this skill out of universal capability
> lists and move it with its owning research tools into the dedicated Claude
> adapter. Use `obsidian-agent:source-digest` for portable plain-note research.
Preserve continuity from research question through an evidence-backed outline.
**REQUIRED SUB-SKILL:** obsidian-agent:vault-grounding
## Process
1. **Project** — identify the active `research-project`; if none exists, offer
`research_project_create` with a focused question.
2. **Source** — import source material with `research_source_import`; preserve
readable content and its computed fingerprint.
3. **Evidence** — capture an exact excerpt and locator with
`research_evidence_capture`, then use `research_evidence_review` before
treating it as trusted.
4. **Claim** — use `research_claim_create` to keep supporting, challenging, and
contextual relations distinct.
5. **Link** — attach further evidence to an existing claim with
`research_claim_link`, choosing `supports`, `challenges`, or `contextualizes`
deliberately; the relation is not decoration, it is what the audit and the
outline read.
6. **Audit** — run `research_audit`; surface stale fingerprints, missing
locators, broken references, unreviewed evidence, and unsupported claims.
7. **Outline** — use `research_outline_generate`; include only reviewed,
locatable, non-stale evidence as trusted support.
## Hard requirements
- Use research tools instead of hand-editing canonical research records.
- Never present proposed, rejected, stale, disconnected, or unsupported material
as trusted support.
- Ask before material vault writes and report the records created or updated.
- Run only the stages needed for the user's request, then name the next useful
stage.
- If a required tool is unavailable, identify it and stop rather than fabricate
state.
## Common mistakes
- Hand-editing a research record instead of using the `research_*` tool that owns it.
- Treating captured evidence as trusted before `research_evidence_review`.
- Generating an outline without running `research_audit` first, so stale
fingerprints and broken references reach the draft.
- Recording every relation as `supports` because it is the default-looking option.
- Inventing project state when a `research_*` tool is unavailable instead of stopping.

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!