The Agent Skills Directory

[PROMPT_INJECTION]: No attempts to override agent behavior, extract system prompts, or bypass safety guidelines were found. The instructional language is benign and task-oriented.
[DATA_EXFILTRATION]: The skill does not perform network requests or access sensitive local directories such as .ssh or .aws. It restricts its file operations to reading strategy documents and writing review artifacts to the project directory.
[REMOTE_CODE_EXECUTION]: No patterns for downloading, piping, or executing remote scripts (e.g., curl|bash) were detected.
[COMMAND_EXECUTION]: The skill does not invoke shell commands, use sudo, or attempt privilege escalation. It focuses entirely on text analysis and markdown/JSON file generation.
[EXTERNAL_DOWNLOADS]: No external resources, dependencies, or remote files are fetched during the execution of this skill.
[DATA_EXPOSURE]: No hardcoded secrets, API keys, or private tokens were identified in the source files.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data (strategy drafts) in Step 2. While explicit boundary markers are not used, the instructions provide a rigorous, multi-step evaluation framework that guides the agent to extract specific structural elements rather than following free-form instructions embedded in the documents.

strategy-review