| name | boltz-check-status |
| description | List recent Boltz Compute jobs across all five endpoints, or inspect a single job by ID, and optionally pull its results onto disk. TRIGGER when the user asks what Boltz jobs are running, to check status, whether a screen finished, to resume a prior job, to download results for a job ID, to see how far along a job is, or after a session restart where earlier job IDs were lost. |
Workflow
Use this skill to recover state across sessions and to inspect or download results for prior Boltz jobs.
- If the user does not provide a job ID, list recent jobs across all five Boltz resource types and summarize the most relevant rows.
- If the user provides a job ID, inspect that job, report the resource type and current status, and include progress or error details when present.
- If the user asks to recover outputs, run the query wrapper with
--download so results and structures are written locally.
Always Do This
- Use the Python SDK wrapper only:
python scripts/query.py ....
- Expect authentication through
BOLTZ_API_KEY; BOLTZ_COMPUTE_API_KEY also works.
- Prefer the default list mode first when the user has lost track of IDs or resource types.
- When inspecting a specific job, explain whether it is still running, succeeded, failed, or stopped.
- When a failed
structure_and_binding job shows only VALIDATION_ERROR / Request validation failed, explain that this endpoint currently omits field-level details and point the user first to missing polymer modifications: [].
- When
--download is used, tell the user where files were written under $BOLTZ_OUTPUT_DIR/<job_id>/.
Input Normalization
- Use the Boltz ID prefix to infer the resource type when present:
pred_* → prediction
prot_des_* → protein_design_ppi
prot_scr_* → protein_library_screen_ppi
sm_des_* → boltz_sm_design
sm_scr_* → boltz_sm_screen
- If the prefix is unfamiliar, the wrapper falls back to probing all supported resources.
- Use
--workspace-id only when the user explicitly needs a non-default workspace and has the right key scope.
- Use
--json only when structured machine-readable output is more useful than the default text table.
Read This Reference When Needed
- Read references/api.md for:
- the exact SDK methods covered by this status skill
- list mode parameters and returned columns
- inspect mode behavior, downloadable artifacts, and result pagination
- details about which resources expose
list_results()
Command Pattern
List recent jobs:
python scripts/query.py
Inspect one job:
python scripts/query.py --id JOB_ID
Inspect and recover outputs:
python scripts/query.py --id JOB_ID --download
Outputs
- Default mode prints a merged recent-job view.
- Inspect mode prints JSON for the located job.
- Download mode writes recovered artifacts under
$BOLTZ_OUTPUT_DIR/<job_id>/.