sec_edgar

Files (1)

• SKILL.md

SKILL.md (first lines)

# SEC EDGAR

Use this skill to answer questions that require SEC EDGAR public company filing
or financial data.

## Setup

Check whether the agent-gw Python SDK is available in the current Python environment, and install it only if the check fails:

```bash
python3 -c "import agent_gw" || python3 -m pip install "$(curl -s https://cdn.kimi.com/agentgw/pysdk/manifest.json | python3 -c "import json,sys; print(json.load(sys.stdin)['latest']['url'])")"
```

The SDK needs an API key from `api_key=...`, `KIMI_API_KEY`, or
`~/.kimi/agent-gw.json`.

## Workflow

1. Run `python3 scripts/sec_edgar_tool.py describe` from the plugin directory to
   call `get_data_source_desc({"name": "sec_edgar"})`.
2. Read the returned Markdown carefully. It contains the overall data source
   rules, SEC company and filing formats, global constraints, and each API's
   description, required parameters, optional parameters, defaults, and allowed
   values.
3. Select the API that best matches the user's question, such as company info,
   filings, XBRL facts, financial statements, insider trades, institutional
   holdings, or material company events.
4. Build `params` exactly from the Markdown requirements. Use documented company,
   ticker, CIK, form type, period, date, accession number, and pagination fields
   only when the API supports them.
5. Use `python3 scripts/sec_edgar_tool.py call` to call `call_data_source_tool`.
6. If the call fails, explain the failure reason from the response.
7. If the call succeeds, save any returned files first, then answer using
   `resp.result.assistant`; ignore `resp.result.user` unless display content is
   specifically needed.

## Script

Use the bundled script from the plugin directory:

```bash
python3 scripts/sec_edgar_tool.py describe
```

After reading the Markdown and selecting an API:

```bash
python3 scripts/sec_edgar_tool.py call \
  --api-name "<api name from markdown>" \
  --params-json '{"required_param":"value"}'
```

For larger params, write a JSON object and pass
`--params-file path/to/params.json`.

The script:

- sends `{"name": "sec_edgar"}` to `get_data_source_desc`
- sends `{"data_source_name": "sec_edgar", "api_name": ..., "params": ...}` to
  `call_data_source_tool`
- prints failure messages from `error.user` or `error.assistant`
- saves returned files to each `files[].name` path returned by the data source
- prints the joined `result.assistant` texts on success

Expected `call_data_source_tool` response shape:

```python
{
    "is_success": bool,
    "result": {"user": list[str], "assistant": list[str]} | None,
    "error": {"user": list[str], "assistant": list[str]} | None,
    "files": [{"name": str, "content": str}],
}
```

When files are returned, `name` is the file path or name to write. The path is
usually dictated by the selected API's params in the Markdown docs. If an API
does not need files, the response normally has no files to save.

From serejaris/kimi-skills, MIT.

More from serejaris/kimi-skills