The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The provided validation script scripts/validate-rest-api-translations.php uses the PHP require statement on files located within the target module directory ({module_dir}/Messages/*.php). Since the skill is designed to analyze and optimize potentially untrusted code (Mode 2 and Mode 3), an attacker could place malicious code in a translation file that is executed when the agent runs the recommended post-generation or optimization checks.
Evidence: scripts/validate-rest-api-translations.php lines 68 and 79: $translations = require $file; and $translations = require $singleFileLayout;.
[DYNAMIC_EXECUTION]: The skill instructions and the validation script facilitate the loading and execution of PHP files from computed paths derived from user input or project structure without sufficient isolation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze existing module code for augmentation and optimization, creating an attack surface where malicious instructions embedded in module files (comments, documentation, or string literals) could influence the agent's behavior.
Ingestion points: SKILL.md instructions for Mode 2 (Augment) and Mode 3 (Optimize) direct the agent to read all files in a module directory.
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' instructions for external code.
Capability inventory: The agent possesses Bash, Write, Edit, and Agent tools, which can be leveraged if the agent is misled by injected instructions.
Sanitization: Absent. Untrusted code is processed directly by the LLM and the included PHP validation script.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file contains a line starting with an exclamation mark (! grep -RE ...) which may be interpreted as a load-time command execution on certain agent platforms. However, the command itself (a grep-based validation) is benign and does not appear to involve exfiltration or dangerous parameters.

mikopbx-module