Terminal execution and logs

A terminal executes on your connected computer. Its conversation activity records the command and its outcome; application logs and Job history answer different questions. Start with connection and permissions and the CLI commands.

Validate an integration from a workspace

In the Asterion sample, a developer checks a quote before changing the portal. The approved workspace contains the application, and Local Access permits its local HTTP service on a compatible runner. The task asks for two quote requests and fixed assertions, without placing an order.

Use the approved Asterion workspace to check the quote service.
Quote AST-FILTER-100 at quantities 4 and 10.
Assert totals of 4625 and 11563 EUR minor units, and reserved=false.
Show the actual exit code and the observed values.

The verified execution returned:

Quantity	Total in minor units	Display total	Stock reserved
4	4625	EUR 46.25	No
10	11563	EUR 115.63	No

The command exited 0 after both assertions passed. Each response had its own Execution ID. This confirms the quote contract for those inputs; it does not confirm an order, invoice, or webhook delivery.

Terminal activity showing two quote responses, totals 4625 and 11563, and exit code zero — Actual terminal command, captured output and exit code from the synthetic quote service.

Read the activity

Expand the work summary and the terminal action in the conversation. Inspect the selected terminal/workspace, command, captured output and completion state. A long-running command can remain active while the Assistant continues with other steps. Use its existing execution to inspect progress, send input where supported, or cancel it.

Evidence	What it establishes	What to check next
Command accepted	The runner accepted the action	Wait for its terminal state
stdout/stderr	What the process emitted	Check truncation and exit status
Exit code 0	The program reported success	Confirm the assertion or durable business result
Nonzero exit	The program reported failure	Read the relevant output before retrying
Timeout or cancellation	Execution reached a stopping boundary	Inspect partial effects; do not assume rollback
Truncated output	The displayed evidence is incomplete	Use a narrower command or a small result file

If a connection disappears after a write, recover the existing command status and inspect the destination before executing it again. A repeated shell command can repeat external effects.

Other useful workflows

Task	Suitable command outcome	Additional verification
Run a repository test suite	Named tests and exit status	Review skipped tests and coverage of the changed behavior
Inspect a supplier CSV	Header, row count and a few sanitized rows	Validate types and rejected rows before importing
Generate an invoice preview	Local file and renderer success	Open the file; upload and finalize separately if needed
Debug a browser flow	Live page state and observed totals	Use Live Preview for the actual controlled browser
Exercise an Endpoint	HTTP status, expected result, Execution ID	Follow any separately returned Job or delivery ID

Avoid printing environment dumps, credentials, cookies or signed URLs. Commands can print sensitive content even when the command itself looks harmless.

Distinguish three log surfaces

Surface	Contents	Audience and use
Conversation terminal activity	Command, captured output and execution state	Authorized conversation viewers; explain what the Assistant actually ran
CLI terminal operational log	UTC time, severity, correlation identifiers and selected lifecycle metadata	Runner owner; diagnose connection, policy, timeout and cancellation problems
Component application logs	Messages deliberately emitted with `api.log`	Authorized instance operators; investigate business execution

The CLI operational log lives under the CLI configuration directory at logs/terminal/<deviceId>/debug.log. It records selected metadata such as exit code, timeout, truncation and browser action; it does not copy raw shell commands, stdout or stderr into that file. The conversation’s captured output is a separate surface. Background service process logs are separate again. Operational logs rotate daily, compress old files and retain archives for 30 days. Files are owner-readable/writable and directories are owner-only. Log rotation is not a substitute for keeping secrets out of command output.

Write a useful application log

For a quote Component, record counts and totals instead of the whole request:

api.log({
  message: 'Contract quote calculated',
  args: { lineCount: 1, totalMinor: 4625, currency: 'EUR' },
}, 'INFO');

Native tests show these diagnostic entries beside the corresponding case. For a running Endpoint or Job, use its Execution ID and Observability. A log message is diagnostic evidence; it cannot replace a passing assertion or a finalized Storage receipt.

Start here

Build and integrate

Operate

AI and Agents

Help

Terminal execution and logs

Validate an integration from a workspace

Read the activity

Other useful workflows

Distinguish three log surfaces

Write a useful application log

​Validate an integration from a workspace

​Read the activity

​Other useful workflows

​Distinguish three log surfaces

​Write a useful application log

Validate an integration from a workspace

Read the activity

Other useful workflows

Distinguish three log surfaces

Write a useful application log