The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands, including package management via npm install, document conversion using soffice.py (LibreOffice) and pdftoppm, and content extraction through markitdown for quality assurance.
[EXTERNAL_DOWNLOADS]: The workflow requires downloading well-known Node.js packages (simple-icons, sharp, pptxgenjs) from the public NPM registry to support logo processing and slide generation.
[REMOTE_CODE_EXECUTION]: The skill utilizes a dynamic execution pattern where it generates a JavaScript script at runtime using the pptxgenjs library and executes it to produce the final document. This script incorporates data retrieved from external API sources.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests untrusted data from external financial records and processes it in a pipeline that involves reading the final output back into the agent's context.
Ingestion points: Data enters the context via S&P Global tools such as get_info_from_identifiers and get_rounds_of_funding_info_from_transaction_ids as described in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the external data interpolated into the slide or script.
Capability inventory: The agent utilizes shell access (npm, python), file system access to /home/claude/ and /mnt/user-data/, and the ability to execute generated scripts.
Sanitization: No sanitization or validation logic is defined for strings retrieved from external APIs before they are included in the generated PowerPoint script or text fields.

fsi-spg-funding-digest