The Agent Skills Directory

[SAFE]: The automated detection of remote code execution via curl | python is a false positive. The instruction curl -s http://localhost:8080/health | python -m json.tool uses the standard Python json.tool module to pretty-print JSON data, which is a safe formatting operation that does not execute the input as code.
[INDIRECT_PROMPT_INJECTION]: The skill triages incidents by processing application logs (/var/log/app/app.log) and API responses. This creates a potential surface for indirect prompt injection if an attacker can control log output to influence the agent's interpretation.
Ingestion points: File access to /var/log/app/app.log via grep and HTTP responses via curl.
Boundary markers: None explicitly defined for log ingestion.
Capability inventory: Includes Bash(curl) for health checks and python for authentication verification.
Sanitization: The skill counts occurrences of error patterns in logs but does not perform specific sanitization of the log content before processing.
[SAFE]: Handling of sensitive credentials follows security best practices. The skill checks for the presence and metadata (length) of the FOUNDRY_TOKEN environment variable without printing the secret value to the console or logs.

palantir-incident-runbook