A Claude skill that digs undocumented institutional knowledge out of a company's record — Slack, email, docs, tickets, CRM, code, meeting notes — and returns an answer where every claim carries a link back to where it came from.
Anything that can't produce a link doesn't ship as a finding. It ships as a gap.
That rule is the point. A knowledge document's value is that the next person can trust it without re-verifying. An unsourced claim looks exactly as authoritative as a sourced one, gets copied into the next doc, and becomes fake institutional knowledge that outlives everyone who could have corrected it. A gap tells the reader "nobody knows this, go find out" — which is true and actionable.
Reach for it when someone asks:
- Why does our org do this a particular way?
- What was decided about X, and why?
- What does a new hire or a handoff need to know?
- What only lives in one person's head?
- Is this written down anywhere?
- Has this come up before? (precedent search)
It also fits the case where someone says "search our Slack for X" but what they actually need is a sourced answer rather than a pile of search hits.
git clone https://github.com/TenexTony/tribal-knowledge.git ~/.claude/skills/tribal-knowledgeRun /reload-plugins, or it loads automatically next session.
Download a zip of this repo and upload it through the skills UI:
gh repo archive TenexTony/tribal-knowledge --format zip --output tribal-knowledge.zipIf you build the zip by hand, compress the repo contents — SKILL.md and
.claude-plugin/ must sit at the top level of the archive, not inside a nested
folder. Zipping the folder itself is the most common reason a skill fails
validation.
Five phases, described in full in SKILL.md:
- Inventory the record — check which connectors are actually live before asking anything, and say plainly what's missing.
- Interview — three questions at most, skipped entirely for specific asks. Establishes the outcome and the scope.
- Hunt for traces, not answers — the trail is in the exceptions, reversals, and repeated questions, not in a document that states the policy.
- Verify — resolve every claim to a citable location, or demote it to a gap.
- Deliver — the artifact the outcome called for, with gaps listed as gaps.
There is also a Sensitive material section covering what to do when the hunt turns up personnel matters, compensation, or legal exposure.
Supporting references:
| File | What's in it |
|---|---|
references/sources.md |
What lives where and how to search it — chat, email, wikis, trackers, CRM, code hosts, meeting notes |
references/hunts.md |
Hunts organised by function and by outcome |
references/traces.md |
Seven trace patterns with query templates — the repeated question, the exception, the reversal, the name, the workaround, the warning, the stale artifact |
The skill is only as good as what it can reach. It needs at least one connected source — a chat platform, email, a doc store, a ticket tracker, a code host, or an uploaded export. With nothing connected it will tell you so rather than guess, which is the correct behaviour but not a useful session.
MIT — see LICENSE.