The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes utility scripts like scripts/ddmin.py and scripts/bisect_oracle.py to run software testing oracles. These scripts are implemented using subprocess.run with list-based arguments and shlex.split for safe command parsing, effectively preventing shell injection.
[DYNAMIC_EXECUTION]: Documentation and tests demonstrate the generation and execution of temporary test scripts (e.g., check.py) to verify software behavior. These are standard practices for debugging and are performed within a local, controlled context.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as failing inputs and repository histories. Although this creates an attack surface for indirect prompt injection, the tools treat these inputs as data for minimization and analysis.
Ingestion points: Ingestion of failing input files (e.g., in scripts/ddmin.py) and repository history (during git bisect).
Boundary markers: The workflow documentation encourages the use of strict oracles and clear failure signatures to avoid drift.
Capability inventory: The skill utilizes subprocess.run for command execution, git for history traversal, and standard file system operations.
Sanitization: Oracle commands are tokenized using shlex.split, and inputs are passed to subprocesses via temporary file paths.

debug-software-failures