The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a user-provided shell command ('METRIC_COMMAND') within an autonomous loop. Once the setup is complete, the agent runs this command repeatedly without pausing for review, which allows for the continuous execution of arbitrary logic on the host system.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection through its ingestion of untrusted data. 1. Ingestion points: The agent reads the contents of repository files in 'IN_SCOPE_FILES' and the output of the metric command in 'run.log'. 2. Boundary markers: The instructions lack delimiters or warnings to ignore embedded instructions within the ingested content. 3. Capability inventory: The agent has the ability to modify files, perform git operations, and execute shell commands. 4. Sanitization: No sanitization or validation of the content read from the repository or logs is performed before it is used in the 'THINK' phase.
[COMMAND_EXECUTION]: The autonomous 'never pause' instruction removes human-in-the-loop oversight for code modifications and command execution, increasing the risk of unintended or harmful changes to the environment.

autoresearch