The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to build and run software services (S0/S1), install command-line interfaces (CLIs), and execute build workflows. These operations involve running arbitrary commands that could be malicious if the build environment or instructions are compromised.
[DYNAMIC_EXECUTION]: Step 3 of the 'Reproduce' section requires the agent to 'Create and execute a portable PoC' against live deployments. This involves the generation and execution of code at runtime based on the provided regression candidate.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (preparation packages, candidates, and registries) which could contain malicious instructions designed to exploit the agent's high-capability environment.
Ingestion points: Input data enters through 'sealed preparation/discovery packages' and 'candidate and registry assignments' referenced in the 'When to Use' and 'Reproduce' sections.
Boundary markers: The skill mentions following a 'file contract' and 'security guidance', but does not specify technical delimiters or 'ignore' instructions for the processed content.
Capability inventory: High-risk capabilities including building/installing software, running network services, and executing PoC code are present in the 'Reproduce' section.
Sanitization: While output is 'sanitized' (Step 4), there is no mention of sanitizing or validating the input data before processing or execution.
[EXTERNAL_DOWNLOADS]: The instruction to 'Build/install CLIs and libraries' (Step 2) implies the installation of external dependencies, which may be fetched from remote registries at runtime.

reproduce-regression