mk:confluence-spec-analyst
Forks to the confluence-spec-analyst agent. Read-only at the Confluence side — the agent has Write tool privilege ONLY for persisting the Spec Research Report locally; it never writes to Confluence itself.
Triggers
- "analyze spec for 12345"
- "extract requirements from this confluence page"
- "spec research on https://acme.atlassian.net/wiki/spaces/ENG/pages/12345/Q3-Roadmap"
- "what are the gaps in PAGE-ID 12345"
Operations
analyze PAGE-ID [--include-children N] [--no-images] [--with-commands] — --with-commands is opt-in (default report has suggestions only, no executable commands).
See also
- Agent:
../../agents/confluence-spec-analyst.md - Shared:
../confluence/references/{install-and-auth,cli-idioms,safety-framework}.md - Domain refs:
references/spec-analysis-patterns.md— gap-detection heuristics + weasel-word inventoryreferences/report-template.md— the Spec Research Report skeletonreferences/integration-with-planning-engine.md— how planning-engine consumes the report
- Peer leaves:
mk:confluence-page(raw CRUD),mk:confluence-search(CQL search),mk:planning-engine(consumes this report via--specflag)
Handoff
mk:planning-engineaccepts--spec <report-path>pointing at this report. User runs spec-analyst FIRST, then passes the resulting report path to planning-engine.mk:intakerecognizes Confluence URL/page-id as a 5th source — fetches raw page content via wrapper, recommends spec-analyst for deeper analysis.mk:jira-issue create— human runs manually from the report's Suggested Commands (with--with-commandsflag).
Gotchas
- Source page hash detects staleness; v1 produces a fresh report on every re-run (no diff mode).
mk:multimodalabsent → Image / Diagram Findings section is omitted with[NO_MULTIMODAL]flag; rest of analysis proceeds.- Children pipeline: hard cap of 10 children. Default depth 1. User must explicitly raise both.
--with-commandsis opt-in — default report has suggestions only; preserves "reports not automation".- Grow this list as new edge cases surface.