Document agent-watcher

[dragon-ai-agent]

Context

agent-watcher now runs scheduled reports and trace refreshes for watched curation repositories. We should document it in aidocs so people know what it does, where outputs live, and how to operate it.

Repo: https://github.com/ai4curation/agent-watcher

Suggested content

• Purpose: scheduled neutral activity digests plus qualitative Claude summaries for watched repos.
• Watched targets and cadence: config/targets.json; report issues use agent-watcher, report, and target:<short_name> labels.
• Main workflows: scan-watched-repos.yml, review-agentic-setup.yml, refresh-public-traces.yml, and validate.yml.
• Public trace bundle: public-traces/, organized by GitHub owner/repo and retrieval surface; covers Actions traces, Dragon PR traces, and optional Copilot logs.
• Operations: dispatch refresh/report workflows, tune LIMIT_* inputs, add/remove watched repos, and manage required tokens/secrets.
• Caveats/security: traces can include full model messages and tool output; run secret scans before publishing; large traces may be gzipped and duplicate payloads deduplicated.

Acceptance criteria

• Add an aidocs page that links to the agent-watcher README and design docs.
• Include a short operator checklist for adding a target and interpreting report/trace issues.
• Include the current trace coverage expectations for DisMech and ai-gene-review, including repo-specific Dragon/scanner/shepherd workflows.