The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's hooks section (pre_execution, post_execution, on_error) utilizes npx claude-flow@alpha. This command automatically fetches the specified package from the public NPM registry at runtime.
[REMOTE_CODE_EXECUTION]: The use of npx to run claude-flow@alpha constitutes remote code execution, as the package is downloaded and executed in the user's environment without explicit version pinning or integrity verification.
[DATA_EXFILTRATION]: In the post_execution hook, the skill transmits $TASK_OUTPUT (the generated API code) and $TASK_CONTEXT to the claude-flow tool via the memory store-pattern and neural train commands. This sends potentially sensitive intellectual property to an external utility.
[COMMAND_EXECUTION]: The shell scripts in the hooks section interpolate variables like $TASK and $TASK_CONTEXT directly into shell commands. This creates a risk of command injection if these variables contain shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill implements a 'self-learning' mechanism that ingests existing project files.
Ingestion points: find . -name "*.route.js" -o -name "*.controller.js" in the pre_execution hook.
Boundary markers: None present to distinguish between instructions and data.
Capability inventory: Bash, Write, Edit, Read, MultiEdit, Task tools are all enabled.
Sanitization: No visible sanitization or validation of the ingested files before they are processed by the learning protocol.

agent-dev-backend-api