Initialize Skene Configuration

Set up Skene automatically. The user runs /skene-init, everything gets configured, and they see a one-line summary. Zero questions unless something is genuinely missing and cannot be inferred.

Config reference: Configuration guide.

Design principle

Make init invisible. Detect everything from existing config, env vars, and user-level config. Only ask the user if the agent truly cannot figure it out (e.g. API key missing everywhere).

Steps

1. Discover CLI options Run uvx skene config --help to see what flags are available. Use only flags that appear in the output.

2. Check what already exists (run in parallel where possible):

   • ls .skene.config 2>/dev/null — project config
   • ls ~/.config/skene/config 2>/dev/null — user-level config
   • echo n | uvx skene config --show — full resolved config (piped to avoid blocking)
   • printenv SKENE_API_KEY SKENE_PROVIDER 2>/dev/null — env vars

3. Decide what to do (no user interaction needed):

   Case A — Config exists and is valid (provider set, API key present, provider/model match): Done. Report summary.

   Case B — Config exists but has issues (provider/model mismatch, missing API key, etc.): Fix what you can automatically:

   • Provider/model mismatch → update .skene.config with correct defaults.
   • API key missing from file but SKENE_API_KEY in env → fine, note it.
   • API key missing everywhere → ask user: "Set SKENE_API_KEY in your environment, then re-run /skene-init."

   Case C — No config at all:

   • Run uvx skene config --init (check --help first to confirm this flag exists).
   • If that fails with permission error, write .skene.config TOML directly.
   • Infer provider from env vars or default to openai.
   • Validate with echo n | uvx skene config --show.

4. Final validation

   echo n | uvx skene config --show
   

   Confirm provider, model, API key. If good, report.

5. Report result "Skene init complete. Provider: X, model: Y. Run /skene-analyze next."

Terminal rules

The user CANNOT type into agent-spawned terminals. The agent handles ALL input.

• Always pipe echo n to uvx skene config --show to skip the edit prompt.
• NEVER run uvx skene config (no flags) — multi-step interactive. Write .skene.config directly.
• Stuck terminal: kill the process and re-run with piped input.
• Genuine blockers only: API key missing everywhere is the only thing that requires user action.

Error handling

• Permission error: Write .skene.config manually. If uvx itself can't run, tell user to run pip install uv in their terminal.
• API key missing everywhere: Ask user to set SKENE_API_KEY.
• Provider/model mismatch: Fix automatically.

Output

One line: "Skene init complete. Provider: X, model: Y. Run /skene-analyze next."