DrugFlow Skills

Route requests to the correct DrugFlow API flow and execute with minimal ambiguity.

Flow Selection

1. Read references/index.md first.
2. Match user intent to one flow.
3. Load only that flow's reference files.
4. Prefer script execution from scripts/<flow>/ when available.

Current Flows

1. Common APIs: reusable auth/workspace/balance/jobs APIs available.
2. Virtual screening: complete flow available.
3. Docking: complete flow available.
4. ADMET: complete flow available.
5. Rescoring: complete flow available.
6. Structure extract: complete flow available (img2mol backend type).
7. Molecular factory: complete flow available (with atom-selection helpers).

Common APIs Workflow

1. Read references/flows/common-apis/call-flow.md.
2. Read references/flows/common-apis/payloads.md.
3. Reuse scripts/common/drugflow_api.py for:

• signin
• signup
• list_workspaces / create_workspace / ensure_workspace
• get_balance
• list_jobs

4. Use scripts/common/test_common_apis.py for direct smoke tests.

Virtual Screening Workflow

1. Read references/flows/virtual-screening/call-flow.md.
2. Read references/flows/virtual-screening/payloads.md.
3. Use scripts/virtual-screening/run_vs_flow.py for end-to-end execution.
4. Always include ws_id for /api/jobs list/detail.
5. For /api/jobs create, pass name, type, args (JSON string), ws_id; in non-private mode include expect_tokens and avail_tokens.

Docking Workflow

1. Read references/flows/docking/call-flow.md.
2. Read references/flows/docking/payloads.md.
3. Use scripts/docking/run_docking_flow.py for end-to-end execution.
4. Create docking jobs through POST /api/jobs with multipart fields pdb, ligands, pdb_content, and args.
5. Site-driven docking box note: when --site is provided but center/size/radius are omitted, the script auto-derives the docking box from that site in local PDB.
6. Always include ws_id on job list/detail requests and pass expect_tokens/avail_tokens in non-private mode.

ADMET Workflow

1. Read references/flows/admet/call-flow.md.
2. Read references/flows/admet/payloads.md.
3. Use scripts/admet/run_admet_flow.py for end-to-end execution.
4. ADMET job type is fixed to admet-dl.
5. Support two input modes:

• direct smiles list
• dataset mode via dataset_id + smiles_col

6. For /api/jobs create, pass name, type=admet-dl, args, ws_id, and in non-private mode expect_tokens/avail_tokens.

Rescoring Workflow

1. Read references/flows/rescoring/call-flow.md.
2. Read references/flows/rescoring/payloads.md.
3. Use scripts/rescoring/run_rescoring_flow.py for end-to-end execution.
4. Create rescoring jobs through POST /api/jobs with:

• type=rescoring
• form fields pdb, ligands, smiles_col
• args.mode=semi and args.rescoring_functions

5. Script enforces input files: --pdb-file must be .pdb, --ligands-file must be .sdf.
6. Always include ws_id; in non-private mode include expect_tokens and avail_tokens.

Structure Extract Workflow

1. Read references/flows/structure-extract/call-flow.md.
2. Read references/flows/structure-extract/payloads.md.
3. Use scripts/structure-extract/run_structure_extract_flow.py for end-to-end execution.
4. User-facing "结构提取" maps to backend job type=img2mol.
5. For create, pass name, type=img2mol, args (dataset_id, page_list), ws_id, and in non-private mode expect_tokens/avail_tokens.
6. dataset_id must be img2mol-compatible and include extras.osskey.

Molecular Factory Workflow

1. Read references/flows/molecular-factory/call-flow.md.
2. Read references/flows/molecular-factory/payloads.md.
3. Use scripts/molecular-factory/run_molecular_factory_flow.py:

• atom-info
• extract-partial
• draw-atom-index
• create-job

4. Default to non-docking molecular factory unless user explicitly asks for docking:

• args.need_docking=false
• args.pdb_use.*=false

5. Default generation models:

• args.molgen_algos=["Frag-GPT","REINVENT"]

6. Use helper APIs first to confirm selected_atoms/start_atoms, then submit molecular_factory job.
7. Always pass ws_id; in non-private mode include expect_tokens and avail_tokens.

Output Contract

1. Return method + endpoint + required parameters for each step.
2. Return key ids and state: ws_id, job_id, state, result count.
3. When running scripts, return command + important outputs.

Expansion Rules

1. Add new flow docs under references/flows/<flow>/ with call-flow.md and payloads.md.
2. Add runnable scripts under scripts/<flow>/.
3. Update references/index.md and this file's Current Flows section.

References

1. references/index.md
2. references/flows/common-apis/call-flow.md
3. references/flows/common-apis/payloads.md
4. references/flows/virtual-screening/call-flow.md
5. references/flows/virtual-screening/payloads.md
6. references/flows/docking/call-flow.md
7. references/flows/docking/payloads.md
8. references/flows/admet/call-flow.md
9. references/flows/admet/payloads.md
10. references/flows/molecular-factory/call-flow.md
11. references/flows/molecular-factory/payloads.md
12. references/flows/rescoring/call-flow.md
13. references/flows/rescoring/payloads.md
14. references/flows/structure-extract/call-flow.md
15. references/flows/structure-extract/payloads.md