The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/exhaustive_shrink.py executes shell commands provided as the 'oracle' to verify code changes.
Evidence: In scripts/exhaustive_shrink.py, the run_oracle function uses subprocess.run(command, shell=True) to execute the behavior-preservation command specified in the --oracle argument.
[INDIRECT_PROMPT_INJECTION]: The skill processes entire external repositories, which creates a potential surface for instructions embedded in data (such as code comments or documentation) to influence the agent.
Ingestion points: scripts/code_slim_audit.py and scripts/architecture_dx_scan.py ingest all code and text files within a target repository for analysis.
Boundary markers: The skill instructions establish behavior preservation and user-approval gates as logical boundaries, though it lacks explicit technical delimiters for raw file content.
Capability inventory: The skill has the capability to write to the filesystem, delete files, and execute shell commands via the oracle script.
Sanitization: scripts/exhaustive_shrink.py includes an assert_safe_relpath function that validates candidate paths to prevent directory traversal attacks (e.g., preventing the use of .. to access files outside the repository root).

exhaustive-code-slimmer