The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to ingest and execute 'Approved Implementation Plans' and 'Research Documents' from external phases.
Ingestion points: Requirements are read from Phase 1/2 documents.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the input plans are provided.
Capability inventory: The skill performs file writes, git commits, and executes system commands via cargo test, cargo clippy, and cargo audit.
Sanitization: No validation or sanitization of the external plan content is performed before it is used to generate and execute code. A malicious plan could embed instructions that result in unauthorized file access or data exfiltration during the 'Verification' loop.
Dynamic Execution (HIGH): The core workflow involves the agent writing code and unit tests based on external input and then executing them.
Evidence: The 'Implementation Workflow' and 'Test-First Implementation' sections explicitly instruct the agent to write code to pass tests and then run cargo test. If the plan specifies a malicious test case, the agent will execute that arbitrary code on the host system.
Command Execution (MEDIUM): The skill makes extensive use of subprocess calls to tools like git and cargo.
Evidence: Instructions include git commit, cargo test, cargo clippy, and cargo audit. While these are standard development tools, they are executed based on the state of the code generated from external plans.

disciplined-implementation