---

name: reporting description: "Generates, monitors, reads, chats with, manages, and troubleshoots MiroFish simulation reports and report-side graph tools." disable-model-invocation: true metadata: disco-role: operating license: AGPL 3.0

reporting

Use this sub-skill when a task asks to generate a MiroFish report from a completed simulation, follow report progress, read section_01.md, inspect agent_log.jsonl or console_log.txt, download or delete a report, chat with the Report Agent, diagnose report returns 409, understand Step 4/5 report interaction, or reason about insight_forge / panorama_search / report graph-tool behavior.

Route first

• If the graph is not built, stale, being deleted, or needs Zep ingestion work, route to the sibling graph-build sub-skill.
• If the simulation has not been prepared, started, stopped, or finalized, route to the sibling simulation-run sub-skill before attempting report generation.
• If the user wants Step 5 direct interviews or surveys with simulated platform agents rather than Report Agent chat, route those simulation interview calls to simulation-run; keep only Report Agent chat here.

Read or run the bundled material

• Read references/workflows.md when generating a report, polling logs/sections/progress, moving from Step 4 to Step 5, chatting with the Report Agent, or using report-side debug tools.
• Read references/api-reference.md when constructing /api/report/* requests, interpreting response payloads, or deciding what an HTTP status code means.
• Read references/report-artifacts.md when inspecting report directories, progress.json, outline.json, meta.json, section_*.md, full_report.md, agent_log.jsonl, or console_log.txt.
• Read references/troubleshooting.md when report generation is blocked, stale, missing logs, missing graph data, failing in a report tool, or affected by fabricated <tool_result> tags.
• Run python scripts/report-log-summary.py --help to inspect the bundled report-directory summarizer, or python scripts/report-log-summary.py --self-test to smoke-check it without MiroFish services.

Minimal safe operating loop

1. Start with either a completed simulation_id or an existing report_id. Do not generate a report while simulation execution or Zep graph-memory ingestion is active.
2. For a new report, call POST /api/report/generate with JSON { "simulation_id": "...", "force_regenerate": false } or true; force_regenerate must be a JSON boolean.
3. Store both report_id and task_id from the response. The report directory and log endpoints can be polled immediately after the request returns.
4. Poll GET /api/report/{report_id}/agent-log?from_line=N and GET /api/report/{report_id}/console-log?from_line=N; optionally poll GET /api/report/{report_id}/progress and GET /api/report/{report_id}/sections for file-backed state.
5. Treat planning_complete as the outline boundary, section_complete as the durable section boundary, and report_complete plus progress.status == "completed" as final completion.
6. After completion, read or download full_report.md, use GET /api/report/check/{simulation_id} to confirm interaction unlock state, and use POST /api/report/chat for Report Agent follow-up.