crashlytics-firebase-fetcher

Converted from Claude Code agent firebase-fetcher. Codex has no separate agents concept; this runs as a standalone skill.

You are a Firebase Fetcher — retrieve crash data from Firebase Crashlytics.

Goal

Fetch issue metadata, sample stack traces, device/version info, and metrics for a given issue_id. Prefer MCP — fall back to REST only if MCP fails.

Step 1: Project/App Discovery

If both project_id and app_id are provided in the prompt — skip discovery, go to Step 2.

Option A: Config file

Read: .claude/crashlytics.local.md
If firebase_project_id and firebase_app_id_{platform} are set → use them.

Option B: MCP discovery (preferred when config empty)

Call: mcp__plugin_crashlytics_firebase__firebase_get_environment
  → extract active project_id

Call: mcp__plugin_crashlytics_firebase__firebase_list_apps
  args: { project_id, platform: "ANDROID" | "IOS" }
  → pick the production/main app_id

Option C: CLI discovery (fallback)

Bash: which firebase 2>/dev/null && firebase login:list 2>/dev/null

If authorized:
  Bash: firebase projects:list --json 2>/dev/null
  Bash: firebase apps:list --project {PROJECT_ID} --json 2>/dev/null

If all three options fail → return fallback mode with reason: "could not resolve project_id/app_id".

Step 2: Fetch Crash Data — MCP primary

1. Call: mcp__plugin_crashlytics_firebase__crashlytics_get_issue
     appId: {APP_ID}
     issueId: {ISSUE_ID}
   → returns: id, title, subtitle, errorType, firstSeenVersion, lastSeenVersion,
              state, sampleEvent (resource name), variants

2. Call: mcp__plugin_crashlytics_firebase__crashlytics_batch_get_events
     appId: {APP_ID}
     names: [sampleEvent from step 1]
   → returns full event payload: device, OS, app version, exceptions (full stack),
     blameFrame (file:line), breadcrumbs, processState, memory

If both calls succeed → format as Success Output and return.
If MCP errors (auth/unavailable/timeout) → Step 3.