agentsclimarketplace

Knowledge base governance

Skill kpince/pi-knowledge-cognition/skills/knowledge-base-governance

Pi skills for recognizing knowledge-base opportunities and deciding what belongs

Install
npx -y skills add kpince/pi-knowledge-cognition --skill knowledge-base-governance

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Govern proposed knowledge shared among Pi agents through human validation, immutable correction, supersession, contradiction, freshness, permissions, and reviewed promotion. Use when agent-local findings may become shared or canonical knowledge.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.8 KB, 673 tokens by cl100k_base, as published. Nobody here has run it

Knowledge Base Governance

Purpose

Allow useful knowledge to travel between Pi agents without synchronizing raw memory or allowing agents to manufacture shared truth.

Core rule:

Agents may propose. Authorized reviewers promote. History remains inspectable.

States

A backend may use different names, but preserve these meanings:

  • proposed — candidate awaiting review;
  • validated — accepted by an authorized reviewer;
  • corrected — accepted replacement with explicit history;
  • rejected — reviewed and not accepted;
  • deferred — decision postponed;
  • superseded — replaced by a newer version;
  • stale — retained but unsafe for current use without warning;
  • disputed — unresolved contradiction exists.

Promotion Rules

A shared proposal must include:

  • bounded object and scope;
  • proposition;
  • future use;
  • provenance and epistemic status;
  • limitations;
  • sensitivity/access scope;
  • freshness or invalidation conditions;
  • conflicts or related prior knowledge;
  • requested reviewer.

Consequential inferred, estimated, sensitive, or cross-agent claims require human review. Repetition across agents is not validation.

Correction and Supersession

Never silently rewrite a validated proposition.

A correction should:

  1. create a successor;
  2. identify what it replaces;
  3. preserve the previous wording and provenance;
  4. record reviewer, time, reason, and changed scope;
  5. mark dependent conclusions for review where relevant.

Contradictions

When evidence conflicts:

  • retain both evidence references;
  • mark the conflict explicitly;
  • do not choose the newest, most popular, or most confident claim automatically;
  • identify differences in time, scope, method, and authority;
  • request clarification when the conflict affects future action.

Freshness

Freshness is contextual. Record either:

  • a time-based review expectation;
  • event-based invalidators;
  • not_applicable with rationale;
  • unknown when no honest rule exists.

Stale knowledge may remain historically useful but must not silently support a current consequential recommendation.

Sharing Boundaries

Use the narrowest applicable scope:

  • session;
  • agent/profile;
  • project;
  • bounded object/business;
  • selected agents;
  • owner-wide;
  • public/reusable doctrine.

Never promote secrets, credentials, raw sessions, private client data, logs, or unrelated profile state. Do not assume that owner-wide means every runtime should receive the content in its prompt.

Cross-Agent Retrieval

Return bounded, task-relevant knowledge rather than dumping the complete store. Include compact provenance, validation state, scope, and freshness. Exclude rejected, forbidden, unrelated, and unauthorized material by default.

Reusable Learning

A local lesson that may improve the package itself should be proposed as a cognitive change, not silently added to doctrine. State source context, evidence, fitness environment, removed private details, overfit risk, intended scope, and reviewer decision.

Completion Check

  • Promotion authority is explicit.
  • Shared knowledge is not raw memory synchronization.
  • Corrections preserve history.
  • Contradictions remain visible until resolved.
  • Freshness is represented honestly.
  • Retrieval respects task and access scope.
  • Local lessons do not become global doctrine automatically.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.