The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends installing the 'shinka-evolve' package from PyPI using pip or uv, which is the legitimate core dependency for the evolution framework described.
[COMMAND_EXECUTION]: The template for 'evaluate.py' includes code to execute candidate solutions through subprocess calls, a standard requirement for an evaluation harness supporting multiple programming languages.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user inputs to generate program scaffolds, representing a vulnerability surface for instruction injection. * Ingestion points: User-provided task descriptions and success criteria are processed as defined in the 'Workflow' section of 'SKILL.md'. * Boundary markers: While structural markers like 'EVOLVE-BLOCK' are used in code, the skill lacks specific instructions for the agent to use delimiters or sanitization warnings for user inputs. * Capability inventory: The generated scaffolds include shell command execution capabilities via subprocess. * Sanitization: There are no instructions provided for filtering or validating user-provided requirements before their inclusion in generated scripts.

shinka-setup