AI Skill Report Card

Curating Knowledge Governance

A-85·Sep 27, 2026·Source: Web

Knowledge Curator (Knowledge Governance)

14 / 15

Given a knowledge entry (article, FAQ, doc snippet, wiki page), produce three outputs: Knowledge Assessment, Quality Report, Governance Recommendation.

Input: New KB article "How to reset VPN access" submitted by IT team.

Knowledge Assessment:
- Topic: VPN access reset procedure
- Scope: IT support, internal employees
- Freshness: Written today, references current VPN vendor (valid)
- Ownership: IT Support Team (clear owner assigned)

Quality Report:
- Accuracy: Steps match current admin console (verified against screenshot)
- Completeness: Missing rollback step if reset fails — FLAG
- Clarity: Clear, numbered steps, appropriate for target audience
- Duplicate Check: 87% similarity to existing article "VPN Password Reset Guide" (KB-0231) — FLAG

Governance Recommendation:
- Action: MERGE with KB-0231
- Reason: Overlapping content; consolidating avoids fragmentation and conflicting instructions
- Required fix before publish: Add rollback/troubleshooting step
- Priority: Medium
Recommendation▾
Add an edge case example showing a Reject scenario to round out coverage of all governance actions
14 / 15

Progress:

  • Step 1: Intake — identify entry type, source, intended audience, owner
  • Step 2: Assess — evaluate scope, freshness, and factual grounding
  • Step 3: Score quality — accuracy, completeness, clarity, structure
  • Step 4: Duplicate/conflict scan — compare against existing KB entries by topic and semantic similarity
  • Step 5: Decide governance action — publish / revise / merge / archive / reject
  • Step 6: Output the three-part report

Step 1: Intake

Capture: title/topic, source/author, date, target audience, claimed purpose. If any is missing, note as a gap rather than guessing.

Step 2: Assess

  • Scope: Is the topic well-bounded, or does it try to cover too much?
  • Freshness: Does it reference outdated tools, policies, versions, or dates?
  • Ownership: Is there a clear accountable owner/team for future updates?

Step 3: Score quality

Rate each on a simple scale (Pass / Flag / Fail) with a one-line justification:

  • Accuracy — factually correct, verifiable, no contradictions with authoritative sources
  • Completeness — covers prerequisites, steps, edge cases, and what to do if it fails
  • Clarity — unambiguous, appropriate reading level, well-structured (headings/steps/lists)
  • Structure/Format — follows KB template conventions (title, tags, metadata)

Step 4: Duplicate/conflict scan

  • Search existing knowledge base for entries on the same or adjacent topic.
  • Estimate similarity (rough %) based on overlapping steps/claims, not just title match.
  • Distinguish duplicate (redundant, same info) from conflict (different/contradictory info on same topic) — conflicts are higher severity.

Step 5: Governance decision

Choose one action:

  • Publish — passes all checks, no duplicates/conflicts
  • Revise — quality gaps must be fixed first (list exactly what)
  • Merge — significant overlap with existing entry; consolidate
  • Archive — outdated/superseded, keep for historical record only
  • Reject — inaccurate, out of scope, or unsalvageable

Step 6: Output

Always structure the final answer under the three fixed headings below.

Recommendation▾
Include guidance on handling ambiguous ownership when multiple teams could plausibly own an entry
Knowledge Assessment:
- Topic:
- Scope:
- Freshness:
- Ownership:

Quality Report:
- Accuracy: [Pass/Flag/Fail] — justification
- Completeness: [Pass/Flag/Fail] — justification
- Clarity: [Pass/Flag/Fail] — justification
- Duplicate Check: [None/Duplicate/Conflict found] — details + % similarity if applicable

Governance Recommendation:
- Action: [Publish/Revise/Merge/Archive/Reject]
- Reason:
- Required fix before publish (if any):
- Priority: [Low/Medium/High]
18 / 20

Example 1: Input: Two KB articles both titled differently but both describe "How to submit an expense report," with one saying reimbursement takes 5 business days and the other saying 10.

Output:

Knowledge Assessment:
- Topic: Expense report submission process
- Scope: Finance/employee self-service
- Freshness: One article dated 2022, other 2024
- Ownership: Finance team (unclear which article is canonical)

Quality Report:
- Accuracy: Fail — contradictory reimbursement timelines (5 vs 10 days)
- Completeness: Pass — both cover full submission steps
- Clarity: Pass — both well-written
- Duplicate Check: Conflict found — ~90% overlapping content, contradictory data point

Governance Recommendation:
- Action: Merge
- Reason: Same process described twice with conflicting facts; must consolidate into single source of truth
- Required fix before publish: Confirm current reimbursement SLA with Finance owner, update the surviving article, archive the other
- Priority: High

Example 2: Input: A well-written onboarding guide with no duplicates found, last updated last week.

Output:

Knowledge Assessment:
- Topic: New hire onboarding checklist
- Scope: HR, clearly bounded to first-week tasks
- Freshness: Updated 7 days ago, current
- Ownership: HR Ops (named owner in metadata)

Quality Report:
- Accuracy: Pass
- Completeness: Pass — includes prerequisites and escalation contact
- Clarity: Pass
- Duplicate Check: None found

Governance Recommendation:
- Action: Publish
- Reason: Meets all quality bars, no conflicts
- Required fix before publish: None
- Priority: Low
Recommendation▾
Consider adding a brief note on how to handle very large knowledge bases where full duplicate scanning isn't feasible (e.g., sampling or tagging strategy)
  • Always check for conflicts, not just duplicates — contradictory facts are more dangerous than redundant ones.
  • Flag missing ownership even if content quality is high; unowned knowledge decays.
  • Prefer merge over reject when content is accurate but overlapping — preserve institutional knowledge.
  • Treat freshness relative to the domain: security/API docs decay fast; policy/process docs decay slower.
  • Keep the governance action singular and decisive — avoid vague recommendations like "review further."
  • Don't approve content solely because it's well-written — accuracy and duplication checks are mandatory regardless of prose quality.
  • Don't call something a "duplicate" just because titles are similar; verify actual content overlap.
  • Don't skip the "required fix" field when recommending Revise or Merge — vague recommendations block action.
  • Don't assign "Publish" if ownership is missing — unowned content should be at minimum flagged in the report.
0
Grade A-AI Skill Framework
Scorecard
Criteria Breakdown
Quick Start
14/15
Workflow
14/15
Examples
18/20
Completeness
18/20
Format
14/15
Conciseness
14/15