Narrow skill description to match When Not to Use

[illmethinks0]

What

The SKILL.md frontmatter description still reads:

Always use browser-harness for any web interaction: automation, scraping, testing, or site/app work.

0.1.7 added a When Not to Use section that says the opposite for the simple cases - prefer curl for a plain HTTP fetch of public content. The description was never updated to match. git log -L 3,3:SKILL.md shows it unchanged apart from the browser-use -> browser-harness rename.

Why it matters

In Claude Code and Codex the frontmatter description is the always-loaded field, and it is what decides whether the skill gets invoked at all. The body only loads once the skill fires. So the always-visible text tells agents to always use the browser for any web interaction, while the correction sits in the half that may never be read.

The practical effect is agents opening a browser for tasks the skill itself says should use curl.

Change

One line. Keeps trigger coverage for genuine browser work (interaction, logged-in sessions, JS-rendered or bot-protected pages) and drops the claim over plain fetches.

Related to #554 but independent - that one is daemon connection classification, this is skill metadata.

---

Summary by cubic

Update SKILL.md description to match guidance: use browser-harness for real-browser interactions (clicking, typing, navigation, logged-in, JS-rendered or bot-protected), not plain HTTP fetches. Prefer curl for simple public fetches to avoid unnecessary browser launches.

^{Written for commit 06cbbc7. Summary will update on new commits.}

[browser-harness-review]

✅ Skill review passed

Reviewed 1 file(s) — no findings.

[cubic-dev-ai]

No issues found across 1 file

_{Re-trigger cubic}

[Kastan97]

Ollie (CTO) Review: APPROVED ✅

What this does: Narrows skill description to align with the existing When Not to Use section. Matching docs to intended scope.

Good to merge once CI passes.