The Agent Skills Directory

[COMMAND_EXECUTION]: Several utility scripts execute shell commands provided via arguments to run the repository's test or gate suites.
scripts/redteam_runner.py uses subprocess.run to execute a user-specified command via the --gate-cmd flag.
scripts/manual_mutation_runner.py uses subprocess.run with shell=True to execute a command provided by the --test-cmd argument.
scripts/fail_to_pass_check.sh is a bash script that uses bash -c to run a test command passed as a script argument.
[DYNAMIC_EXECUTION]: The skill performs dynamic modification of the codebase to verify test suite effectiveness, followed by the execution of that code.
scripts/manual_mutation_runner.py reads source files, applies string-replacement 'mutants', writes the modified file to disk, runs tests, and then restores the original content.
scripts/redteam_runner.py applies git patches to a temporary worktree and executes the gate command against the modified code.
[INDIRECT_PROMPT_INJECTION]: The skill contains multiple ingestion points for external, potentially untrusted repository data, which it processes to audit quality gates.
Ingestion points: scripts/assess_repo.py (crawls files and reads manifests), scripts/banned_apis_check.py (scans core logic files), scripts/spec_trace_check.py (reads SPEC.md and test files), and scripts/test_weakening_check.py (analyzes git diffs).
Boundary markers: No explicit markers or delimiters are used in the script logic to separate analyzed data from agent instructions.
Capability inventory: The skill has the ability to read and write files and execute shell commands via the scripts listed above.
Sanitization: No explicit sanitization or filtering of the ingested content was observed in the scripts.

agent-proof-codebase