The Agent Skills Directory

[OBFUSCATION]: The validator script scripts/validate_skill.py defines sensitive sandbox paths using string concatenation (e.g., "/home/" + "cl" + "aude"). While this is used to define a blacklist for a security validator without triggering static analysis scanners on the validator itself, it is a technique used to evade pattern-based detection.
[DYNAMIC_EXECUTION]: The file scripts/validate_skill.py utilizes the Python compile() function to verify the syntax of bundled scripts within a skill package. Furthermore, scripts/test_validate_skill.py uses importlib to dynamically load and execute the validator module for unit testing. These are standard local execution patterns for quality assurance in this development context.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a generator that processes user-supplied capability descriptions and existing skill files to create or repair skill packages, establishing an ingestion surface for potentially untrusted data.
Ingestion points: User-provided $ARGUMENTS and existing skill folder content (markdown and scripts) read during the audit and creation procedures in SKILL.md.
Boundary markers: No explicit delimiters or warnings are instructed for separating user input from the generated instructions.
Capability inventory: The skill is designed to write files to the local system and execute local Python-based validation and auditing scripts via the command line.
Sanitization: The skill includes a structural and syntax validator (scripts/validate_skill.py) that checks for sandbox path leaks and compilation errors in the generated content.
[COMMAND_EXECUTION]: The SKILL.md instructions guide the agent to execute several local Python scripts for validating skill primitives and package structures, which is expected behavior for a developer-oriented tool within its local repository environment.

skill-creator