The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's configuration templates (e.g., for .claude/settings.json) interpolate tool parameters like ${tool.params.command} and ${tool.params.file_path} directly into shell command strings. This creates a high risk of command injection if the input values contain shell metacharacters such as backticks or semicolons that could execute unintended code.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install an external package from the npm registry: claude-flow@alpha.
[PROMPT_INJECTION]: The features for 'neural pattern training' and 'memory coordination' ingest operational data (like code edits) to influence future agent behavior, creating a surface for indirect prompt injection. 1. Ingestion points: post-edit --train-patterns, pre-task --load-memory. 2. Boundary markers: Absent. 3. Capability inventory: Shell execution, file modification, and tool coordination. 4. Sanitization: Absent.
[DATA_EXFILTRATION]: The documentation provides examples of accessing and processing sensitive files, such as production.env. While presented as a safety backup, accessing credential-heavy files increases the potential impact of other vulnerabilities and provides a vector for data exposure.

Hooks Automation