The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/script_tester.py utility uses subprocess.run to execute Python scripts found within the target skill's directory. This is the intended functionality of the meta-skill, designed to verify script syntax, imports, and runtime behavior. The SKILL.md file contains a clear warning: 'WARNING — runs the target's code. ... Never point them at an untrusted or unreviewed skill; run skill-security-auditor on it first.'
[SAFE]: Detections for eval and exec patterns in references/quality-scoring-rubric.md and tests/test_security_scorer.py are confirmed false positives. In the rubric, these appear as educational examples of insecure coding practices to avoid. In the test suite, they are string literals used to verify that the security detection regexes in scripts/security_scorer.py function correctly.
[SAFE]: The skill maintains a strict policy of using only the Python standard library, which is verified by its own _find_external_imports logic. This reduces the risk of supply chain attacks through third-party dependencies.
[INDIRECT_PROMPT_INJECTION]: The skill processes external file data (SKILL.md, scripts, assets) from target directories. The analysis confirms these ingestion points are used for structural validation, code quality scoring, and testing. There is no evidence of this content being unsafely interpolated into agent prompts to hijack the tester's logic.

skill-tester