knowledge-base-governance
Skill0 starsChanged 2 months agoMITView on GitHub
Use when creating a knowledge_base/ entry or deciding whether an existing one should move between lifecycle states. Governs the three-state lifecycle — emergent, validated, canonical — which is deliberately lightweight: no scheduled review, no forced promote-or-archive deadline.
In this repository16 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- anti-ai-writingSkill
- installSkill
- librarianSkill
- project-creationSkill
- services-catalogSkill
- session-teardownSkill
- anti-ai-writingSkill
- installSkill
- knowledge-base-governanceSkill
- librarianSkill
- project-creationSkill
- services-catalogSkill
- session-teardownSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 4 are distinctive, and 3 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
Every knowledge_base/ entry carries a status.
- Very specific: line 29
Write it up from .agents/templates/knowledge_base_node_template.md.
- Distinctive: line 36
When you use a concept in real work, go back and update it:
- Very specific: line 40
update date either way
- Distinctive: line 42
Do it at the moment of use, while you still remember.
- Distinctive: line 52
Keep the field accurate all the same.
- Distinctive: line 58
No validation path: If nothing would count as evidence, the entry can never legitimately move.
≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Knowledge Base Governance
Every knowledge_base/ entry carries a status. It records one thing: how much evidence there is that this idea actually works.
That distinction is what makes the base worth drawing on. A thought you had once on a call and a framework you have used with five clients are not the same kind of knowledge, and a base that flattens them cannot be trusted when you reach into it.
The three states
emergent — new. Extracted from a call, an article, or a moment of synthesis. Untested, possibly wrong, possibly about to be rewritten. Every entry starts here, without exception.
validated — used once, and it held up. You put it into real work and it did what you expected. You would reach for it again.
canonical — used repeatedly, roughly three times or more. It has become part of how you think, and people who follow your work recognise it as yours.
emergent ──[used once, and it worked]──► validated ──[used 2+ more times]──► canonicalMovement is earned by use, never asserted at creation. Creating something directly as validated because it feels solid defeats the whole mechanism: the status stops meaning "tested" and starts meaning "I liked it", at which point you may as well delete the field.
Creating an entry
- Extract the concept from its source — an artifact, a library item, or your own synthesis.
- Write it up from
.agents/templates/knowledge_base_node_template.md. - Set
status: emergent. - State how you would validate it. What would prove this works? What piece of work would test it? An entry with no validation path can never legitimately be promoted, because nothing would count as evidence.
- Link it to at least two related concepts with
[[wiki-links]]. An orphan node is one you will never find again at the moment you need it.
Promoting an entry
When you use a concept in real work, go back and update it:
- first use →
emergentbecomesvalidated - third or later use →
validatedbecomescanonical - update
dateeither way
Do it at the moment of use, while you still remember. Retroactive promotion is guesswork about your own history.
No scheduled review, on purpose
There is no staleness check here, and no deadline by which an emergent entry must be promoted or archived. That is a design decision, not an omission.
Most work gets made organically rather than assembled from a concept you went looking for. So a large and permanent pool of untested emergent entries is the expected steady state, not a backlog demanding attention. Treating it as a queue to clear produces exactly the wrong behaviour — entries get promoted to shorten the list rather than because anything validated them, and within a quarter the status field means nothing at all.
Promote or archive when you notice something concrete: you used it, or it is plainly dead weight. Never on a timer.
Keep the field accurate all the same. It is what lets you ask the base for what has actually held up, rather than for everything you have ever thought.
Anti-patterns
- Creating an entry as
validatedorcanonical. Everything starts emergent. - Promoting to tidy up. The pool of emergent entries is supposed to be large.
- No validation path. If nothing would count as evidence, the entry can never legitimately move.
- Orphan nodes. Two links minimum, or it is invisible when you need it.
- Ten concepts from one source. Concept sprawl degrades the base faster than sparseness does.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

