The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses the dynamic context injection pattern (!command) in the Context section of SKILL.md to identify existing project configuration files. These commands (e.g., find . -maxdepth 1 -name '.editorconfig') are restricted to the local directory, do not incorporate external user input, and are used solely for situational awareness during skill initialization.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential surface for indirect prompt injection as it reads project metadata files like package.json and pyproject.toml to detect development tools. Ingestion points: Step 1 involves reading package and project configuration files. Boundary markers: No explicit delimiters are used for these inputs. Capability inventory: The skill can modify local files using Write, Edit, and Bash tools. Sanitization: The risk is effectively mitigated because the skill uses the detected files only as triggers to apply static, safe templates from REFERENCE.md rather than interpreting or executing the content of those files directly.

configure-editor