The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill incorporates an 'Intake' phase that persists external, untrusted data (such as pasted specifications, schema tables, and API contracts) verbatim into the local project structure at docs/versions/{version_san}/sdd/source/. This data is later processed to generate requirements.md and subsequent implementation task-*.md files. This creates a surface where malicious instructions hidden in the specifications could influence the generated code or sub-agent actions. The skill attempts to mitigate this through mandatory human-in-the-loop checkpoints (Gates 0 through 4).
[COMMAND_EXECUTION]: The skill instructions involve executing the @nextstage-brasil/harness CLI tool using npx for tasks like repository preparation, skill installation, and brownfield mapping. These are vendor-provided tools invoked for their intended purpose.
[DYNAMIC_EXECUTION]: The skill manages a sophisticated orchestration layer that dynamically dispatches execution to specialized sub-agents (e.g., task-writer-agent, coder-agent, reviewer-agent) and can programmatically install missing 'complement' skills (like ns-frontend-design or ns-docs-writer) as needed during the delivery journey.

ns-spec-driven