The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs subagents to execute arbitrary shell commands for compilation and building (e.g., go build ./..., npm run build, cargo build) based on configurations found in project files like tech-stack-rule.md.
[DYNAMIC_EXECUTION]: The subagents are tasked with generating source code and testing suites which are then compiled and run within the host environment. This runtime code generation and execution cycle is a significant security boundary crossing.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided specification files (UCS files) as the primary instruction set for its subagents. If these specification files are sourced from an untrusted party, they could contain malicious instructions designed to exploit the agent's capabilities.
Ingestion points: Ingests files from docs/specs/API-UCS/ (SKILL.md).
Boundary markers: None identified in the prompt templates to distinguish between data and instructions.
Capability inventory: Subagents have capabilities to write files, modify project scripts, and execute shell commands.
Sanitization: No sanitization or validation of the specification content is performed before processing.
[PERSISTENCE]: The skill automates the modification of project lifecycle scripts, such as Makefile or package.json entries (test-unit, test-e2e, test-all). These modifications could be used to insert persistent malicious commands that execute whenever the developer runs standard test or build tasks.

do-api