airmosaic-ai

CSMAR Data Downloader

English

Invocation status: csmar_browser_download is a non-executable browser-workflow reference. Only csmar_process_company_info is enabled in run-tool for an existing company-info ZIP.

Setup

Use Python 3.10+ with the project installed, pandas and openpyxl for local processing. Browser retrieval requires the member’s institutional access and an actually available browser tool; Node.js/Playwright is optional for a selected JavaScript workflow. Keep data/results, non-secret settings and browser state in separate external AIRMOSAIC_ASSETS_DIR, AIRMOSAIC_CONFIG_DIR and AIRMOSAIC_AUTH_DIR roots. No personal configuration is auto-loaded.

Download CSMAR (China Stock Market & Accounting Research) data tables from https://data.csmar.com/ for socioeconomic driver analysis (workflow Layer 1).

Browser download reference: csmar_browser_download

This catalogued reference has execution_supported: false; run-tool rejects it. Use describe-tool csmar_browser_download for metadata, then follow the authorized browser workflow with a tool actually available in the host, or perform it manually. No project browser-launch/login adapter is registered.

For first access, scoped acquisition, validation and the local-processing handoff, follow the first-download guide. Source quality is maintained in enterprise-source methods.

The verified Chrome example records an eight-field, one-record download and saved-file validation on 2026-10-27. Browser acquisition remains separate from the executable local processor.

Use the task-authorized browser session. If launching a project-managed browser, explicitly configure the member’s own profile under ${AIRMOSAIC_AUTH_DIR}/browser_profiles/csmar/; setting a variable does not select it. Host-managed sessions have separate storage, which these variables do not relocate or enforce. Do not export session state.
Let the member authenticate and verify institutional table access. Do not export cookies or automate password entry.
Select the requested table, dates, records and needed fields. Select all fields only when the task requires the complete table; verify the selection and query size before submission.
Submit the query to create a download record, then open that record and use the final local-save action. Keep signed download links and account details local.
Save the export in the authorized data cache and retain its field description. Existing download records are not proof that a file was saved successfully.

CSV/TXT may be alternatives for browser exports, but the processor below specifically expects a ZIP containing an XLSX workbook; other formats require a separate adaptation. Browser operations have their own authorization and do not inherit runner path enforcement.

Core Table: Listed Company Basic Information

Entry point:

https://data.csmar.com/csmar.html#/datacenter/singletable/search?databaseId=176

Key fields to include:

Stock code, stock short name
Statistical cutoff date
Industry name and code (CSRC classification)
Unified social credit code
Chinese full name
Province, city
Main business
Listing status

Registered local processing: csmar_process_company_info

Use the existing downloaded ZIP; this step needs pandas/openpyxl and no browser login. The public parameter example is a flat JSON object:

{"zip":"${AIRMOSAIC_ASSETS_DIR}/data/01_socioeconomic_natural_change/CSMAR_ListedCompanyBasic/v1/company_basic.zip"}

Map that filename to the authorized input, then run from the repository root:

python -m runtime.cli describe-tool csmar_process_company_info
python -m runtime.cli run-tool csmar_process_company_info --project-id example_project --stage 01_socioeconomic_natural_change --params-file config/examples/run-tool/csmar.json --dry-run
python -m runtime.cli run-tool csmar_process_company_info --project-id example_project --stage 01_socioeconomic_natural_change --params-file config/examples/run-tool/csmar.json

The runner supplies out_dir. The script reads the workbook directly inside the ZIP, without extracting next to or modifying the input. Results go under results/<project_id>/<run_id>/01_socioeconomic_natural_change/csmar_process_company_info/:

csmar_target_industry_company_candidates.csv: sector-matched rows.
csmar_target_industry_company_list_latest.csv: latest company records.
csmar_target_industry_company_list_latest.xlsx: both sheets.

The fixed keyword groups cover thermal power, district heating, iron/steel, nonferrous metals, cement, lime, brick/tile, glass and chemicals. The current implementation uses English, case-sensitive keywords and selects the latest record per company; review language, panel period and empty matches. Its 15-column contract is listed in the first-download guide. Unsupported tables and formats need adaptation. Company identity alone establishes neither cost payer nor physical production boundary. This is not a general financial-statement processor.

Verify outputs

After can_execute=true and an empty blockers list, inspect the actual runner status and the three files above. Open both workbook sheets, compare candidate counts and unique companies, check dates and repeated symbols, and review language-dependent/empty matches. match_keywords lists each sector’s configured vocabulary; it is not a record of the exact words observed in a company. Preserve the input ZIP and field description with the validation record. A completed process does not establish sector coverage or legal/physical eligibility.

References

Browser-flow implementation notes: historical endpoint names and interface boundaries.
Query example: a query declaration, not a download API or executable wrapper.

This site is open source. Improve this page.