The Agent Skills Directory

[PROMPT_INJECTION]: Static detectors flagged instructions that direct the agent to omit specific marketplace configuration commands. A manual review of SKILL.md and references/installing-and-updating.md confirms these are platform-specific optimizations for the Codex environment. The instructions prevent redundant manual steps for the default marketplace path, which is discovered implicitly, and do not represent an attempt to hide malicious actions from the user.
[INDIRECT_PROMPT_INJECTION]: The validation script scripts/validate_plugin.py parses SKILL.md files and plugin.json files from local plugin directories. This is an ingestion point for untrusted data. The script uses yaml.safe_load() for YAML parsing and strict JSON decoding, which mitigates injection risks by preventing the execution of arbitrary code during the validation process.
[COMMAND_EXECUTION]: The skill utilizes local Python scripts (create_basic_plugin.py, update_plugin_cachebuster.py, etc.) to perform file system operations. These scripts use standard libraries such as pathlib and json for secure file handling and do not invoke shell commands with unsanitized user input or external data.

plugin-creator