Diagnose setup and recovery

Output contract: Relay requested CLI fields or output faithfully; preserve warnings/errors and only the explanations required below. For suggested actions, pair Shell with the returned Copilot command; honor commandGuidance. If absent, say "Copilot: no verified equivalent"; never invent a slash command.

Boundary: machine-local; no repository or Story required. Use explicit arguments or SFlow-returned paths; never search $HOME or infer a repository.

1. Run singularity-flow workspace current --json. When it returns a repository, use only its exact
   repositoryPath as cwd; when it returns no selected repository, do not search for one.
2. With a repository, run singularity-flow init --check --json, then
   singularity-flow doctor $ARGUMENTS. Report each failure with its exact safe fix and summarize
   warnings separately. The initialization check covers workflow, portfolio, templates, prompts,
   and governed-agent files on the current branch. Recommend /sf-init when assets are missing.
3. Without a repository, run singularity-flow workspace doctor --json. It is machine-local and
   offline by default. Add --network only after the contributor explicitly chooses to contact the
   remotes named by unfinished bootstrap sessions. Relay proxy and certificate configuration source
   names, never their values; never request credentials or recommend disabling TLS.

All checks above are read-only. Do not reset, stash, switch branches, or edit configuration unless
the user explicitly asks you to apply a fix.

When the user asks about monorepo or Git slowness, pass --performance --json.
Relay the measured total/scoped file counts, warm status/fingerprint times, clone
mode, and recommendations. The benchmark is read-only: never enable FSMonitor,
untracked cache, sparse checkout, or a partial-clone fallback on the user's
behalf merely because the report recommends evaluating it.

When a failure is not explained by the current state, read what actually happened with /sf-logs — singularity-flow logs --level warn shows recent failures and refused tool calls, including the reason a hook blocked a command. Diagnose from the log rather than retrying blindly.