The Agent Skills Directory

[SAFE]: No malicious instructions, unauthorized file access, or network communications were found. The skill is an administrative utility for project documentation.\n- [PROMPT_INJECTION]: The skill manages an indirect prompt injection surface by converting conversation history into agent behavioral rules. Evidence chain: (1) Ingestion points: Session conversation history; (2) Boundary markers: The skill requires structured formatting but lacks explicit input delimiters; (3) Capability inventory: Modifies steering files (.cursor/rules, AGENTS.md, CLAUDE.md) that constrain future agent actions; (4) Sanitization: Comprehensive decision logic (Update/Merge/Conflict/Ignore) and a quality checklist requiring functional few-shot examples mitigate the risk of adopting malicious or unintended rules.\n- [DATA_EXFILTRATION]: The skill's operations are confined to local configuration files and do not involve network requests or the handling of sensitive credentials.\n- [COMMAND_EXECUTION]: The skill does not invoke shell commands or execute code. Its output is limited to text content for documentation and configuration markdown files.

code-evolver