feat(skills): add campus and sourceharbor triage skills

[xiaojiou176]

What these skills teach

• campus-copilot-current-view-triage: how an agent turns one Campus Copilot snapshot or MCP-backed current view into a single evidence-backed next action for a student.
• sourceharbor-watchlist-briefing: how an agent inspects one SourceHarbor watchlist, reuses the current briefing/story payload, and answers one operator question with evidence and one concrete next action.

Why this revision is stronger

• both skills now start with what the agent learns
• MCP/host setup is explicit in references/INSTALL.md, references/OPENHANDS_MCP_CONFIG.json, and references/OPENCLAW_MCP_CONFIG.json
• safe-first capability guidance is explicit in references/CAPABILITIES.md
• first-success walkthroughs are explicit in references/DEMO.md
• troubleshooting is explicit in references/TROUBLESHOOTING.md
• the skill language is now centered on agent capability and truthful evidence use, not repo promo copy

Boundaries kept explicit

• Campus stays snapshot/current-view/read-only first
• SourceHarbor stays evidence-backed and does not pretend to be hosted SaaS
• neither skill claims a live OpenHands listing while this PR is still under review

Validation

• local packet folders were refreshed from the source repos and rechecked before this PR refresh

[enyst]

It seems these files don’t exist

[enyst]

Thank for the PR, @xiaojiou176 . Like in the other PR, please see: #150 (review)

[xiaojiou176]

Addressed the self-containment issues in both skills:

• removed the broken repo-local companion references
• added local references/ docs for MCP/HTTP setup, capability maps, and example output
• rewrote both SKILL.md files to be more progressive-disclosure and to explain what the skill teaches the agent, what runtime it needs, and how to use the exposed MCP/HTTP abilities

[enyst]

I still don’t understand what these skills refer to, sorry.

If you wish to recommend them to users, you can create a marketplace.json and define the skills in it, it doesn’t have to be the default marketplace.

[xiaojiou176]

Added another pass focused on the likely reviewer follow-ups:

• Campus skill now includes a local input-shape.md so the snapshot/current-view fallback is self-contained
• SourceHarbor skill now includes a local http-fallback.md with the concrete watchlist/briefing endpoints
• narrowed wording that looked too much like plugin/repo-internal terminology

[xiaojiou176]

Small reviewer guide for the latest revision:

What changed since the earlier draft:

• both skills now keep their companion material inside local references/ instead of pointing at missing repo-local paths
• Campus now includes references/input-shape.md, so the snapshot/current-view fallback is self-contained
• SourceHarbor now includes references/http-fallback.md, so the HTTP-only path has concrete watchlist/briefing endpoints instead of just a base URL
• wording was narrowed to read more like public skill cards and less like internal plugin/repo rollout notes

Fastest places to review:

• skills/campus-copilot-current-view-triage/SKILL.md
• skills/campus-copilot-current-view-triage/references/input-shape.md
• skills/sourceharbor-watchlist-briefing/SKILL.md
• skills/sourceharbor-watchlist-briefing/references/http-fallback.md

Goal of this revision:

• keep the skills progressive-disclosure and public-reusable
• make both fallback paths self-contained enough that the agent does not have to chase missing docs outside the skill package

[xiaojiou176]

Thanks. These two skills are intended to teach very concrete MCP-backed triage workflows.

• campus-copilot-current-view-triage teaches an agent how to start from Campus Copilot snapshot/current-view surfaces, prefer read-only evidence first, and return one trustworthy next action instead of inventing browser truth.
• sourceharbor-watchlist-briefing teaches an agent how to start from one watchlist, load the current briefing/story context, and answer one operator question with evidence and one next action.

The packet shape is meant to be self-contained: install/config snippets, capability map, demo path, and troubleshooting all live inside the skill folder so the reviewer does not need repo-root context to understand what the agent is being taught.

[xiaojiou176]

Revision note for re-review:

• both packets now start with what the agent learns
• MCP/host wiring is explicit in references/INSTALL.md, references/OPENHANDS_MCP_CONFIG.json, and references/OPENCLAW_MCP_CONFIG.json
• safe-first capability order is explicit in references/CAPABILITIES.md
• first-success flow is explicit in references/DEMO.md
• troubleshooting is explicit in references/TROUBLESHOOTING.md
• the wording now stays on agent capability and evidence-backed next actions instead of repo promo language

This refresh is intended to answer the main reviewer question directly: what should an agent know after using these skills, and how does it verify the first successful run?

[enyst]

@xiaojiou176 As explained in #154 (comment) , I think you have a simple alternative if you wish to provide skills for OpenHands users: make a custom marketplace.json and distribute it to people. Thank you for the contribution!