debug-code-task

name	debug-code-task
description	Investigate IntexuraOS code-task executions when the user shares a `dev.intexuraos.cloud` or `intexuraos.cloud` code-task URL, or provides a `task_*` ID and asks to debug, investigate, or understand what happened.

Debug Code Task

Use this skill for code-task investigations backed by Firestore data.

Trigger patterns

• https://dev.intexuraos.cloud/#/code-tasks/task_*
• https://intexuraos.cloud/#/code-tasks/task_*
• task_* plus intent such as debug, investigate, or what went wrong

Workflow

1. Extract the task ID from the URL hash or direct task_* input.
2. Detect the environment from the URL.
   • dev.intexuraos.cloud = dev
   • intexuraos.cloud without dev. = prod
3. Never fetch the SPA URL. The page is hash-routed; the task data lives in Firestore.
4. Confirm the current machine with uname -n.
5. Use the bundled wrapper .codex/skills/debug-code-task/scripts/fetch-task.sh, which resolves the local repo and runs the Firestore fetcher from the checked-out repo.

Task document

Run:

.codex/skills/debug-code-task/scripts/fetch-task.sh <taskId>

The wrapper uses the checked-out repo for firebase-admin module resolution and expects credentials at ~/.config/gcloud/sa-key.json.

Summarize these fields for the user when present:

• id
• status
• linearIssueId
• workerLocation
• agentType
• result.summary
• error

Log lines

Run:

.codex/skills/debug-code-task/scripts/fetch-task.sh <taskId> --logs-only

Use --logs if you want the task document and logs in one call. Logs are ordered by sequence number and printed as [sequence] message.

Present log evidence directly. Do not infer a root cause unless the logs or task document support it.

Optional follow-up logs

Only do this when Firestore logs are not enough or the user asks.

• Read workerLocation from the task document first.
• If the task ran on home-dev and you are elsewhere, use ssh home-dev.
• Orchestrator logs on home-dev:

journalctl -u intexuraos-orchestrator@pbuchman --since "<time>" --until "<time>"

• Docker container logs while the container still exists:

docker logs claude-worker-<taskId>

Derive the time window from the task document timestamps and pad it by about one minute on each side.

Rules

• Never curl or web-fetch the code-task SPA URL.
• Never run the Firestore script from /tmp.
• Never use node -e for commands that may include !.
• Treat .codex/skills/debug-code-task/scripts/fetch-task.sh as the canonical wrapper path.
• Stay factual. Show the task document and logs before offering conclusions.