The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx -y to automatically download and run the @steipete/oracle package from the public NPM registry without user confirmation.
[REMOTE_CODE_EXECUTION]: By invoking npx, the skill executes arbitrary code from a remote package repository at runtime.
[COMMAND_EXECUTION]: The skill performs several shell operations, including directory creation (mkdir -p), file writing via heredocs (cat > ...), and execution of external CLI tools.
[DATA_EXFILTRATION]: The skill is designed to upload the contents of the src/ directory to an external service via the @steipete/oracle tool for planning purposes. This involves sending proprietary or sensitive source code to a third-party endpoint.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8). It reads existing codebase files and incorporates their content into instructions sent to an external LLM (Oracle), creating a risk that malicious code comments or strings could manipulate the planning output.
Ingestion points: Reads application source code from the src/ directory.
Boundary markers: No delimiters or instructions are provided to the planning model to treat ingested code as untrusted data.
Capability inventory: Shell execution via npx, file system writes, and network transmission of data.
Sanitization: No validation or sanitization is performed on the code content before it is interpolated into the planning prompt.

planner