JSON output and automation

Every data-producing CLI command accepts -j, --json and writes a single envelope to stdout:

{ "ok": true,  "data": <payload> }
{ "ok": false, "error": { "message": "...", "code": "<code>" } }

Three properties make this dependable in scripts:

• Stdout is always valid JSON in this mode: spinners and progress messages are suppressed.
• Exit codes are real: 0 on success, non-zero on failure, 130 on Ctrl+C, and device exec exits with the remote command's code. Pipelines with set -e behave correctly.
• Error codes are stable: not_found, unauthorized, timeout, … the full contract is in error codes, and the envelope shapes (including the fan-out summary/results[] form) in the JSON envelope reference.

Field extraction

-f, --field <dot.path> drills into the result so you don't need jq for simple lookups:

thinr device property store-14 information -f hardware.id
thinr product property edge-gateways temperature -f data.celsius

jq recipes

# fail the pipeline on error
thinr device status store-14 --json | jq -e '.ok' >/dev/null

# devices that failed a fleet command
thinr product exec edge-gateways "systemctl is-active app" --json \
  | jq -r '.data.results[] | select(.ok | not) | .device'

# version inventory across the fleet
thinr product exec edge-gateways "thinr-agent --version" --json \
  | jq -r '.data.results[] | "\(.device): \(.stdout | rtrimstr("\n"))"'

Non-interactive environments

Two things change in scripts and CI compared with your terminal:

• Commands that normally prompt for confirmation refuse to run and fail with needs_confirm; pass -y, --yes explicitly.
• Pick the environment per invocation with THINR_PROFILE=<name> (or --profile; see global options), instead of relying on the default profile of whoever runs the job.

A CI example

# GitHub Actions: deploy a config and verify, against the staging profile
- run: npm install -g @thinremote/thinr-cli
- run: |
    thinr product push edge-gateways ./config.yaml /etc/app/config.yaml --json
    thinr product exec edge-gateways "systemctl restart app && systemctl is-active app" --json \
      | jq -e '.data.summary.failed == 0'
  env:
    THINR_PROFILE: staging

The token comes from the profile store; provision the runner once with a pre-issued token (see Install the CLI) or bake ~/.config/thinr-cli/config.json into the runner's secrets.

Beyond shell scripts

When the automation grows past a few pipelines, two better tools exist: playbooks for multi-step procedures with check mode and staged rollouts, and the MCP server when the thing driving the fleet is an AI agent rather than a script.