The Agent Skills Directory

Command Execution: The skill is designed to interact with the system shell using tools such as curl, grep, tail, and bash sleep. While these are necessary for monitoring tasks (e.g., checking HTTP status codes or log markers), they represent a capability to execute commands based on input parameters.
Indirect Prompt Injection Surface: The skill monitors external data sources like log files, URLs, and process outputs. There is a potential consideration where malicious instructions embedded in these external sources could influence the agent's behavior during the polling loop. The skill includes some internal mitigations by advising the use of narrow commands (e.g., grep -c) to limit the amount of untrusted data brought into the context.
Ingestion points: Untrusted data enters the agent context through monitoring commands targeting log files, URLs, and CI status outputs (SKILL.md).
Boundary markers: The skill does not define specific delimiters for the data ingested during the check phase, though it encourages summarizing outcomes rather than dumping full content.
Capability inventory: The skill has access to shell execution (bash, curl, grep, tail, run_in_background) and sub-agent creation (delegate).
Sanitization: No specific sanitization or filtering of the monitored data is explicitly instructed before it is processed by the agent.
Dynamic Sub-Agent Delegation: The skill utilizes a delegation pattern to create sub-agents with dynamically generated instructions. This is a platform-specific feature intended to improve efficiency in large sessions by offloading the polling logic to a clean context.

monitor