The Agent Skills Directory

[SAFE]: The skill consists entirely of Markdown instructions and YAML configuration. There are no scripts (.py, .js, .sh), binaries, or active code components that could execute on a host system.
[DYNAMIC_CONTEXT_INJECTION]: The SKILL.md file mentions the !command syntax (dynamic context injection) and shell injection. However, these are discussed strictly in the context of analyzing token usage and context load. The skill does not contain any !command directives that would execute during load, nor does it instruct the agent to execute dangerous shell commands.
[INDIRECT_PROMPT_INJECTION]: The skill is intended to process other skill files as data for review. While this constitutes an ingestion of untrusted data, the skill itself lacks the capabilities (such as file-writing, network exfiltration, or code execution) required to turn a successful injection into a high-impact attack. The instructions emphasize preserving safety and validation rules during the optimization process.
[DATA_EXPOSURE]: There are no references to sensitive file paths, environment variables, or hardcoded credentials. The skill focuses on metadata and instruction length rather than data processing.

skill-token-overhead-review