incident

Incident RCA Analysis (Thin Wrapper for prism_analyze)

Takes an incident description (text + optional screenshots) as input, runs multi-perspective analysis via prism_analyze MCP pipeline with UX impact perspective injection and language-aware synthesis, producing a developer-facing RCA report directly from the MCP server.

This skill is a thin wrapper that:

1. Collects incident input and resolves ontology scope (user interaction required)
2. Calls prism_analyze with incident-specific config including perspective_injection and language
3. Polls prism_task_status for progress updates
4. Retrieves results via prism_analyze_result when complete
5. Presents the final RCA report to the user

All incident-owned assets must remain under this skill directory:

• skills/incident/SKILL.md
• skills/incident/templates/report.md
• skills/incident/perspectives/ux-impact.json

When this skill runs from Codex, pass adaptor: "codex" in the prism_analyze call. When it runs from Claude Code, pass adaptor: "claude". Do not rely on process-global runtime defaults when the caller already knows the host runtime.

Do not depend on ~/.codex, cloned copies of this skill elsewhere, or repo-external prompt/template paths.

---

Phase 0: Incident Input Collection

This phase performs only input collection — no ontology resolution, no MCP calls.

Step 0.1: Get Incident Description

Extract the incident description from $ARGUMENTS.

• Description provided → store as {INCIDENT_DESCRIPTION}
• No description → AskUserQuestion (header: "Incident", question: "Please describe the incident to analyze. You can include text description and optional screenshot paths.")

Step 0.2: Screenshot Text Extraction

If the description references image/screenshot paths:

1. Verify each file exists via Read
2. For each screenshot, read the file content using the Read tool (which supports multimodal image reading)
3. Incorporate the visual content description as text into {INCIDENT_DESCRIPTION}