The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute the agent-validate run command, which is the primary mechanism for running the validation suite.
[DATA_EXPOSURE]: The agent reads from and writes to the validator_logs/ directory to access execution logs and structured JSON reports containing violation details.
[INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests and acts upon data from external log files, creating a surface for indirect prompt injection.
Ingestion points: Log files and JSON violation reports located in the validator_logs/ directory are read by subagents and the main agent in Step 3 and Step 6.
Boundary markers: The instructions lack explicit markers or warnings to the AI to ignore instructions or commands that might be maliciously embedded within the log data produced by external tools.
Capability inventory: The skill allows the agent to execute Bash commands and dynamically invoke other skills based on the strings extracted from the logs (e.g., the 'Fix Skill' and 'Fix Instructions' sections).
Sanitization: There is no evidence of sanitization or validation of the 'Fix Skill' names or 'Fix Instructions' extracted from the logs before the agent attempts to execute them.

validator-run