Audit a shared skill library for overlap

Why use this

1. Two skills can sound similar while serving different callers, permissions or output contracts. Combining them solely to reduce library size can remove a specialized safeguard or make routing less predictable.

2. This job could produce an owner-reviewable overlap map with concrete examples. Maintainers could decide whether to merge, narrow or leave skills separate based on conflicting instructions and actual responsibility boundaries rather than titles alone.

Why an agent / simpler alternative

Useful judgment

Compare instruction semantics, triggers and output expectations across a scoped library.

Simpler alternative

Use a naming and ownership inventory when the problem is discoverability rather than conflicting behaviour.

Unsuitable when

The library includes protected instructions or the requester lacks authority to propose changes.

Proposed inputs

1. Name

   Owned skill snapshots

   Owner

   Library maintainer

   Minimum scope

   Authorized versions, descriptions, triggers and owners

   Quality check

   Exclude protected material and identify the current release revision.

2. Name

   Usage and maintenance criteria

   Owner

   Skill owners

   Minimum scope

   Representative tasks, required safeguards and known conflicts

   Quality check

   Distinguish intentional specialization from accidental duplication.

Access and setup — proposed or unknown

Proposed minimum

1. Read authorized skill copies and prepare diffs or consolidation proposals without applying them.

Compatibility unknowns

1. Unknown: repository access, publishing workflow and skill-format compatibility.

External actions requiring approval

1. Editing shared instructions, changing routing or releasing consolidated skills.

Expected deliverable and review checks

Proposed artifact

Overlap audit with evidence pairs, conflict examples, owner assignments and proposed dispositions.

Critical checks

1. Each merge proposal preserves distinct permissions and output contracts.

2. A representative task demonstrates the overlap rather than relying only on similar wording.

Missing output evidence

No new run or output inspection in this editorial layer; see linked source-specific limits.

Failure signals and proposed responses

1. Signal

   A specialized approval rule disappears in a merged draft.

   Why it matters

   Consolidation could broaden action authority unintentionally.

   Proposed response

   List preserved safeguards explicitly and require owner review of the proposed routing.

2. Signal

   Two similar descriptions hide different output formats.

   Why it matters

   A shared replacement may break downstream consumers.

   Proposed response

   Compare callers and schemas; keep separate variants where contracts differ.

Effort and cost — unknown

Setup work

1. Map owners, triggers and output contracts.

Ongoing review

1. Review disputed overlap and maintain a rollback path for approved changes.

Unknowns

1. Unknown: library scale, owner availability and release-validation effort.

First check — proposed, unexecuted

Status

proposed_unexecuted

Synthetic fixture

Synthetic skills share a topic but differ in external-action permission and required output schema.

Acceptance checks

1. The audit recommends separation or a carefully bounded common component.

2. Neither permission nor schema is silently discarded.

Stop conditions

1. Protected content, unapproved edits or a release action would be required.

Limits

1. Less duplication is not automatically better design; preserved contracts and predictable task routing matter more than skill count.