knowledgelisted
Install: claude install-skill jwogrady/spark
# knowledge — minimal decision documentation
`knowledge` captures what a team knows so it stays reusable for humans and AI
agents. It takes raw input — notes, transcripts, architecture decisions, repo
findings, Claude Code session summaries — and files it as clean, durable,
searchable docs. It is a cross-cutting tool, not a lifecycle stage: it runs
whenever [`codify`](../codify/SKILL.md), [`validate`](../validate/SKILL.md), or
[`ship`](../ship/SKILL.md) turn up a decision, system, or process worth keeping,
or alongside [`ideate`](../ideate/SKILL.md) when raw intent needs structure.
(Public docs and positioning are the spark-docs companion plugin's job.)
## The one rule
**Capture truth; mark uncertainty.** Knowledge never invents company facts. It
keeps **facts separate from assumptions**, **current state separate from
intended state**, and flags what it doesn't know instead of smoothing it over. A
clean doc that hides an unknown is worse than a rough one that names it.
## The crew
Three real subagents under [`agents/knowledge/`](../../agents/knowledge/)
(registered as `spark:knowledge:<name>`), orchestrated by this skill:
- **00 Intake** — structures raw input into a fact base and recommends a doc
type. *Barrier.*
- **01 Author** — the one specialist: drafts with the template the doc type
calls for (ADR, system doc, product spec, SOP/runbook) from
[`references/templates.md`](references/templates.md).
- **02 Librarian-Editor** — the lead: placement, dedup, cross-li