Add scope blocks to skills with overlapping boundaries

[szabosteve]

Description

It can be an alternative of #76

During hands-on testing, the skill content was accurate and complete once the agent reached the right file. The main friction was navigation: the agent followed the wrong branch because it was not immediately clear what a skill covered or did not cover.

Adding a short Covers / Does not cover block near the top of relevant SKILL.md files might help agents confirm they are in the right skill before reading the full content, or quickly rule out the wrong branch.

Proposed change

Add a compact scope block to skills where the boundary with neighboring skills is not obvious.

1. Place the scope block near the top of the skill.
2. Use user-facing language where possible.
3. Keep the block short and scannable.
4. Include both positive and negative scope only when it helps routing.
5. Avoid turning the scope block into a full summary.

This should be used especially for:

• hub skills
• broad workflow skills
• skills with overlapping neighboring skills
• skills where agents often choose the wrong branch

Validation

A/B test the selected skills with and without the scope block.

Compare whether the scope block helps agents:

• confirm they are in the right skill faster
• rule out wrong branches earlier
• reduce unnecessary page fetches
• reach the relevant leaf skill with fewer navigation steps
• complete the task without guessing

[szabosteve]

Notes

For example:

## Scope

**Covers:** filtered search, payload indexes, HNSW tuning, quantization, throughput.

**Does not cover:** memory management, index build time.

Slightly different method to make sub-skill descriptions scope-specific, not topic-generic.

Instead of:

Search speed optimization

Something like:

Tuning HNSW parameters, segment count, and quantization for steady-state query latency — not startup or recovery behavior

A negative scope boundary ("not X") is often more useful than a positive one for ruling out fetches.