The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is explicitly designed to retrieve and display sensitive credentials (API keys, OAuth tokens) to the console. This makes these secrets available in the agent's context, where they could be exfiltrated through subsequent agent actions or if the agent is compromised.
[COMMAND_EXECUTION]: Executes a Node.js script (agent-job-secrets.js) that performs authenticated network requests. It uses an AGENT_JOB_TOKEN passed via environment variables to communicate with a dynamic APP_URL.
[DATA_EXFILTRATION]: The AGENT_JOB_TOKEN (a sensitive credential) is sent as an x-api-key header to any URL specified in the APP_URL environment variable. If APP_URL is misconfigured or pointed to a malicious endpoint, this token could be harvested.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input (the secret key name) which is used to construct a network request.
Ingestion points: The get subcommand in agent-job-secrets.js accepts a user-provided key name from the command-line arguments (rest[0]).
Boundary markers: No specific boundary markers or instructions to disregard potential instructions within the input are provided.
Capability inventory: The script performs network GET requests using fetch and outputs the resulting secret value directly to standard output.
Sanitization: The input key is sanitized using encodeURIComponent before being appended to the request URL.

agent-job-secrets