The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes external log files that may contain data from untrusted sources, creating a risk that malicious content could influence the agent's diagnosis and automated responses.
Ingestion points: The agent reads <log_dir>/check_*.log (containing raw output from external commands), <log_dir>/review_*.json (containing violation descriptions), and <log_dir>/.debug.log (event history).
Boundary markers: There are no instructions to use delimiters or ignore potential commands embedded in the ingested log files.
Capability inventory: The skill can execute shell commands via Bash and is instructed to automatically file bug reports using a CLI tool.
Sanitization: The skill does not describe any sanitization or validation process for the data extracted from logs before it is incorporated into a diagnosis or a bug report.
[COMMAND_EXECUTION]: The skill makes extensive use of CLI commands to inspect the local environment and manage state.
It executes the agent-validate utility with subcommands like list, health, detect, validate, and clean (which is destructive) to gather diagnostic evidence.
It uses the which command to locate AI CLI binaries such as claude, gemini, and codex on the system PATH to verify adapter availability.
The instructions direct the agent to invoke the validator-issue tool for automated bug filing under specific conditions, but this tool is not explicitly listed in the allowed-tools section of the YAML frontmatter.

validator-help