adaptorch-cli

Run/evidence JSON may include the SDK's safe, versioned verification_diagnostics. Project reports remain untrusted observations. Invalid optional metadata is omitted by the SDK, including in CLI JSON; it does not turn a valid run response into a CLI error. Existing failed/cancelled/inconclusive exits stay 8/9/10, and a stopped wait still exits 10. PDF tools are not global CLI prerequisites.

Independent AdaptOrch SaaS CLI. Installs the adaptorchctl command and uses adaptorch-client; it does not invoke adaptorch or adaptorch-mcp.

Log in once and every command picks the credential up:

adaptorchctl auth login      # prompts for the key, verifies it, stores it at
                             # ~/.config/adaptorch/config.json (mode 0600)
adaptorchctl auth status     # shows credential_source: env | config | none
adaptorchctl auth logout     # removes the stored key

ADAPTORCH_API_KEY still works and always wins over the stored key, so CI keeps injecting secrets via the environment. config set provider.{name,model, api_key} persists BYOK wiring for run submit; the env vars ADAPTORCH_PROVIDER, ADAPTORCH_PROVIDER_MODEL, and ADAPTORCH_PROVIDER_API_KEY override the stored values. Provider credentials become X-Provider, X-Provider-Model, and X-Provider-Key only on run submit; reads and cancellation never receive the provider key. ADAPTORCH_API_KEY remains the separate tenant key.

Install the current adaptorch-client and adaptorch-cli source packages together. A git push does not replace a previously published PyPI version.

See ../../docs/adaptorchctl-usage.ko.md for usage.

Bounded wait exit codes

adaptorchctl run wait RUN_ID emits the last observation as JSON. A terminal SUCCEEDED exits 0, FAILED exits 8, and CANCELLED exits 9. These codes describe run lifecycle, not result correctness or evidence verification.

A deadline, poll_limit, unsupported_status, or unrecognized stop reason exits 10, even if the last observation looks terminal. Missing or unsupported terminal observations also exit 10. This means the observation is inconclusive, not that the server-side run failed. Inspect reason and retain the JSON; continue with the same run ID when appropriate. Waiting never resubmits or cancels a run.

Automation compatibility: run wait previously returned 0 for these observations. Scripts using set -e must now handle nonzero exits. run get keeps its existing exit-code behavior; submit and cancel still report command success separately.

License

Proprietary — Copyright ClassicMate. All rights reserved. See LICENSE.