The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Bash implementations of wait_for and retry_with_backoff in references/implementation-patterns.md utilize eval to execute arbitrary string conditions and commands. This pattern creates a risk for command injection if the strings are constructed from untrusted or external sources.
[DYNAMIC_EXECUTION]: The Python check_command function in references/implementation-patterns.md employs subprocess.run(command, shell=True). This practice executes a command string directly in a shell environment, which is an unsafe practice that facilitates command injection vulnerabilities when the input is not sanitized.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture creates an attack surface for indirect prompt injection through external data ingestion. 1. Ingestion points: Data enters the system via the HealthCheck.target field (used for URLs, hostnames, and commands) and HTTP Retry-After headers parsed by the RateLimitedClient class in references/implementation-patterns.md. 2. Boundary markers: The provided implementation patterns lack boundary markers or explicit instructions to the agent to ignore potentially malicious embedded content in external responses. 3. Capability inventory: The skill provides functions with capabilities including shell execution (subprocess.run, Bash eval), network requests (requests.get), and network connection testing (socket.connect). 4. Sanitization: There is no evidence of input validation, escaping, or sanitization in the provided code templates, particularly regarding the strings passed to shell-executing functions.

condition-based-waiting