The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to manipulate the subagent's behavior.
Ingestion points: Subagents are instructed to read specifications from docs/specs/ws-UCS/ and docs/specs/ws/, as well as configuration from docs/standards/tech-stack-rule.md (Evidence in templates/ws-impl-prompt.md and templates/ws-test-prompt.md).
Boundary markers: The skill lacks explicit delimiters or instructions for the subagents to ignore potential command-like structures within the read documentation.
Capability inventory: Subagents possess the ability to create and modify source files, and execute shell commands for compilation and testing (Evidence in templates/ws-impl-prompt.md).
Sanitization: There is no evidence of validation or sanitization of the input specifications before they are interpreted by the LLM subagents.
[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands through subagents to perform compilation and testing tasks.
The subagents determine which commands to run by reading docs/standards/tech-stack-rule.md (e.g., test-unit, test-integration, compile).
While these are standard development workflows, the execution path is influenced by the content of external project files which may be untrusted in certain environments.

do-ws