The Agent Skills Directory

[DYNAMIC_EXECUTION]: The instructions direct the agent to generate and execute scripts to verify safety claims about a code change. This involves running code derived from potentially untrusted sources against the local repository. Evidence includes: "you ran a script or test against the real code" and "one small script calling the exact function in question".
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content such as code diffs and external library source code, which could contain hidden instructions meant to compromise the agent's behavior during the analysis phase.
Ingestion points: Code diffs, symbols, and library source files mentioned in SKILL.md Steps 1 and 3.
Boundary markers: None. There are no instructions to delimit or treat the diff content as untrusted data.
Capability inventory: The skill utilizes shell execution and script running to provide proof of safety in SKILL.md Steps 4 and 6.
Sanitization: No sanitization or validation logic is defined for the code being analyzed or executed.
[COMMAND_EXECUTION]: The agent is instructed to run commands and provide the output as part of its reporting process. Evidence: "Report: the key facts, their proof level with the command and output".

ds-blast-radius