Skip to content
Back to skills

Vaultspec Research

ASecurity

Ground a decision in evidence before it is made. Use when an ADR is warranted

  • 2 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 2, 2026
ai-agents

Security analysis

A100/100

Scanned October 2, 2026

npx -y skills add nevenincs/vaultspec-rag --skill vaultspec-research --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Vaultspec Research?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Vaultspec Research
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/nevenincs-vaultspec-research-vaultspec-rag/badge)](https://www.skillsdirectory.com/skills/nevenincs-vaultspec-research-vaultspec-rag)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: vaultspec-research
description: Ground a decision in evidence before it is made. Use when an ADR is warranted
  and the options have not been weighed on evidence already in the vault or the code.
---

# Research (vaultspec-research)

Produces a Research record: the evidence a later ADR decides on. Enter it when the
vaultspec decision test warrants an ADR and the options are not already weighed in the
vault or the code. A question this session can answer without new evidence is answered
in the conversation. Reuse sufficient existing Research, Reference, or Audit evidence
before creating more. This skill terminates within one run.

## Steps

- Ground per the `vaultspec-discovery` rule, decisions first: prior ADRs and research on
  the feature, read whole. Link what already exists; do not restate it.
- Scaffold: `vaultspec-core vault add research --feature {feature}` (or the `create`
  tool). Read `.vaultspec/templates/research.md`; its hint blocks fix the body shape:
  answer-first lead, claim-first `## Findings`, closing `## Sources`.
- Research in this run, or dispatch `vaultspec-researcher` with the question and
  existing evidence. Preserve the returned findings' locators. The ADR persona drafts
  decisions from evidence; do not dispatch it for research-only work.
- When the decision needs grounding in real code, branch to `vaultspec-code-research`
  for a Reference record and link it in `related:`.
- Fill every section this session; never leave or present an unfilled record.
- Verify with `vaultspec-core vault check all`.

## Quality gate

Judged by decision value per token; later phases re-read this record.

- **Answer-first.** The lead states question, stakes, and conclusion; each finding opens
  with its claim.
- **Locator-anchored.** Every non-obvious claim carries a re-fetchable locator (URL,
  `file:line`, commit SHA, `package@version`, RFC); an unanchored claim is marked as
  opinion.
- **Comparative and specific.** Alternatives named with why each was kept or rejected;
  versions, dates, and numbers pinned.
- **Grounding, not deciding.** At most name the option the evidence favours and what the
  ADR must settle. Decisions are recorded only in the ADR.
- **Bounded and lean.** Uninvestigated areas stated; link, do not copy; no hedging, no
  restated prompt, no closing summary.

## Next

Proceed to `vaultspec-adr`, in this run or the next; the ADR's approval covers the
findings, so this record needs no reply of its own. If the user faults the findings,
revise this record's body in place.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.