The Agent Skills Directory

[COMMAND_EXECUTION]: The skill allows the agent to run 'reversible local commands' to inspect the repository and verify safety assumptions during the audit process.
[DYNAMIC_EXECUTION]: The agent is instructed to generate and run 'temporary probes' using the real code and dependency versions to test risks. These probes are created and executed at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the repository environment which may contain hidden instructions that the agent might inadvertently follow.
Ingestion points: The skill reads external content from git diff outputs, commit messages, and source code files during the changed contract identification and consumer search phases.
Boundary markers: The instructions do not define delimiters or specific safety markers to differentiate between the data being audited and instructions for the agent.
Capability inventory: The agent has the capability to execute local shell commands and generate/execute dynamic scripts (probes).
Sanitization: No sanitization, escaping, or validation of the ingested content is specified in the skill's instructions.

blast-radius-audit