The Agent Skills Directory

[DATA_EXFILTRATION]: The Debug Collection Script in SKILL.md accesses the FATHOM_API_KEY environment variable to authenticate requests to api.fathom.ai. It fetches potentially sensitive meeting metadata, including meeting IDs, titles, and timestamps, as well as transcript status information. This data is stored in local files (recent-meetings.json, latest-transcript-status.json) within a diagnostic bundle intended for support purposes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the external Fathom API and instructs the agent to analyze the resulting files, which presents a surface for indirect prompt injection if meeting titles or other metadata contain instructions designed to influence the agent.
Ingestion points: Data enters the system via curl requests to the Fathom API within the Bash script in SKILL.md, saving output to local JSON files.
Boundary markers: The instructions include manual redaction steps for the user, but lack technical delimiters or "ignore embedded instructions" warnings for the agent when it reads the collected diagnostic files.
Capability inventory: The skill is permitted to use Read, Bash (including curl and jq), and Grep to handle and analyze the data.
Sanitization: There is no automated sanitization or filtering of the content retrieved from the API before the agent is prompted to analyze it using jq or grep.

fathom-debug-bundle