diagnose

Diagnose (printed, never transmitted)

A support report the operator can read before anyone else does. This skill PRINTS. It never sends, uploads, posts, mails, or opens an issue, and it has no channel of its own. The operator decides whether any of it is pasted anywhere.

Read-only over everything: no brain file is written, no profile is edited, no connector is re-probed for the sake of this report. The one time it reaches the network is the Update currency section below - a single outbound READ of the latest public version number, never a write and never anything about you. Reading a public release tag is not transmitting your report.

Print this line first, before any collected value:

Nothing below has been transmitted. organic-os collected it locally and printed it here. Read it, remove anything you would rather not share, then decide whether to paste it into an issue.

What to collect

Resolve the brain: registry.get_active() when running interactively; a scheduled run receives the brain path from the routine configuration. Every item below is best-effort - a section that cannot be read prints unknown (<one-line reason>) and the report continues. Missing information is stated, never guessed.

1. Runtime - platform.platform() and platform.python_version(), plus which surface this is (Claude Code CLI or Cowork).
2. PyYAML - present or absent. Try the import; report the outcome, not a path or a version guess.
3. Plugin version - the version field in plugin/.claude-plugin/plugin.json.
4. Brain schema - core.contracts.check_schema(brain_path): the version, whether it is compatible, and the action string it returns.
5. Registry - core.registry.load(): how many sites are registered and which slug is active. The active site's URL is fine to print; the other sites' URLs are not. Print the count, not the list.
6. Connectors - the profile's connectors: block, one line each: the name, the status (verified / unavailable / declined), the context the probe ran in (local-cli, cowork-cloud, ci), and the date it was checked. The context is the load-bearing part - a connector verified from an interactive session is not a claim about the runtime that runs the routines. Plus one row in the same format for the search-data adapter, read from this session's own tool list rather than the profile: name (search-data adapter (openseo) - the known adapter), status present or absent in this session, context this-session, today's date, and - when present - which of the known tools were seen (the known surface is listed in plugin/docs/connectors.md, OpenSEO section; report the tools actually seen, never the list). Looking at the session's tool list is not a re-probe: nothing is called and nothing transmitted.
7. Setup scorecard - the newest runs/*-setup-scorecard/REPORT.md in the brain, if one exists: its date and the per-check status column. Say "no scorecard recorded" when there is none.
8. Last routine outcome - the newest runs/<date>-*/REPORT.md and its date, plus the tail of the runtime log if one exists (~/.config/organic-os/routine-<name>.log): whether the last run finished, and when.
9. Recent error - the most recent error line from that log or from the last run report, if any. One line, quoted as found.