The Agent Skills Directory

[SAFE]: The skill provides a purely instructional methodology for architectural decision-making and does not contain any malicious code or instructions. It utilizes a manual-first invocation posture, ensuring the agent only acts upon explicit user requests.
[COMMAND_EXECUTION]: Documentation and README files include shell command examples (e.g., uv run python, npm test) for local validation and development tasks. These are intended for the skill developer or maintainer and are not executed automatically as part of the agent's runtime workflow.
[PROMPT_INJECTION]: As the skill processes user-supplied architecture problems and can perform external research (Phase 5), it is theoretically susceptible to indirect prompt injection. However, the methodology's focus on structured framing and explicit evaluation criteria serves as a natural defense against malicious input.
Ingestion points: User prompts describing architecture problems and external data from browsing or research tasks.
Boundary markers: Utilizes structured markdown templates and defined output headers for decision artifacts.
Capability inventory: Granted shell, network, filesystemRead, and filesystemWrite permissions to support artifact generation and research activities.
Sanitization: No explicit sanitization or filtering logic is described for ingested data.

future-aware-architecture