The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for processing data from SEC filings (10-K and 10-Q), which are external sources that could contain malicious text designed to influence the agent's behavior.
Ingestion points: Data is extracted via the get_sec_filing tool as described in references/sec-filings.md and SKILL.md Step 2.
Boundary markers: The skill lacks explicit boundary markers or instructions to isolate the ingested text from the agent's primary control flow, increasing the risk that instructions hidden in filings could be obeyed.
Capability inventory: The skill performs file operations using openpyxl and executes shell commands/Python scripts to audit and recalculate workbooks.
Sanitization: While the skill emphasizes financial normalization (sign, scale, and currency), it does not specify any sanitization of the text content pulled from the filings to prevent prompt injection.
[COMMAND_EXECUTION]: The skill instructs the agent to execute several local utility scripts via the command line to manage the Excel build process.
Evidence: SKILL.md references the use of python .agents/skills/xlsx/scripts/recalc.py, audit.py, and insert.py to validate and modify workbooks.
[DYNAMIC_EXECUTION]: The skill follows a pattern where the agent is instructed to generate its own code and then execute it.
Evidence: SKILL.md specifies: "Build as a saved Python script (for example /build_model.py)" and provides instructions to run this generated script to construct the model.

3-statements