The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves and processes conversation transcripts from the Gong API, which are external and untrusted data sources that could contain malicious instructions.
Ingestion points: Transcripts are fetched via curl commands in SKILL.md and the transcript command in scripts/gong.sh.
Boundary markers: The SKILL.md file contains a "Safety Boundaries" section providing instructions to the agent, but no technical delimiters or escaping mechanisms are used for the transcript data itself.
Capability inventory: The skill has permissions to read local configuration files (~/.config/gong/credentials.json) and perform network operations via curl.
Sanitization: There is no evidence of sanitization or filtering of transcript content before it is passed to the LLM.
[COMMAND_EXECUTION]: The helper script scripts/gong.sh is vulnerable to parameter and command injection if the agent passes unvalidated input to the script's arguments.
Evidence: The script uses shell variables $1 (for call_id) and $DAYS directly inside shell commands like date and within JSON strings passed to curl -d. For example, date -d "$DAYS days ago" and "{\"filter\":{\"callIds\":[\"$1\"]}}" perform no validation on the input, potentially allowing shell metacharacters to alter command execution or JSON structure.

gong