The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the external IndyKite REST API (GET /whoami), specifically the type and id fields. The risk of the agent interpreting these strings as instructions is mitigated by explicit boundary markers in SKILL.md and references/whoami-reference.md, which instruct the agent to treat them strictly as plain identifier strings and never as commands to follow. Evidence chain:
- Ingestion points: API response fields in SKILL.md and whoami-reference.md.
- Boundary markers: Clear warnings in the 'Read the response' section of SKILL.md and the 'Response' section of whoami-reference.md.
- Capability inventory: The skill uses curl within a shell script (scripts/whoami.sh) to perform requests.
- Sanitization: The instructions mandate treating input as static data, and the script uses host-pinning for network safety.
[COMMAND_EXECUTION]: The skill uses a local bash script (scripts/whoami.sh) to execute curl commands. This script is implemented with strong security practices, including environment variable checks and strict host-pinning that refuses to send credentials to any domain other than the official IndyKite regional API endpoints (eu.api.indykite.com and us.api.indykite.com).
[DATA_EXFILTRATION]: The skill manages sensitive API keys and bearer tokens. It correctly utilizes environment variables for secret management and includes a redaction mechanism in its script's preview mode (--print) to ensure that live credentials are not inadvertently exposed in logs or console output.

indykite-ciq-whoami