The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes program code and feedback from a SQLite database and interpolates it into a Markdown bundle, creating a potential surface for indirect prompt injection if the resulting bundle is processed by another agent.
Ingestion points: scripts/inspect_best_programs.py loads data from a local database file (defaulting to programs.sqlite) using the load_programs_to_df utility.
Boundary markers: The output code snippets are wrapped in standard Markdown code blocks (e.g., ```python or ```text).
Capability inventory: The script performs file reads and writes (Path.write_text) but does not execute the data it processes.
Sanitization: The script truncates code length based on configuration but does not otherwise sanitize or filter the content for malicious instructions.
[DYNAMIC_EXECUTION]: The script dynamically modifies sys.path to import the shinka utility library from the repository root if it is not already available in the Python environment. This is a common developer pattern for local module resolution.
[COMMAND_EXECUTION]: The skill documentation includes standard shell commands for listing directory contents and executing the Python inspection script, which are appropriate for its function as a developer tool.

shinka-inspect