The Agent Skills Directory

[METADATA_POISONING]: The skill configuration and instructions reference non-existent or hypothetical AI model versions such as gpt-5.6-sol-max and claude-fable-5-1-thinking-max. These references constitute misleading metadata regarding the operational capabilities of the environment.
[INDIRECT_PROMPT_INJECTION]: The workflow relies on ingesting grounding artifacts and design candidates from external skills and model outputs. This ingestion process presents an attack surface where instructions embedded in processed architectural data could influence the agent's behavior during synthesis.
Ingestion points: Grounding artifacts from the ps-how skill and design candidates from the ps-arena process described in Phase B.
Boundary markers: The skill includes a mandatory screening phase against references/design-red-flags.md to identify architectural flaws.
Capability inventory: The agent is instructed to perform file-write operations to implement the chosen architecture in Phase D.
Sanitization: The skill does not define explicit sanitization or filtering of the external architectural content before processing.
[COMMAND_EXECUTION]: The skill instructions direct the agent to modify the project's source code by replacing placeholder 'not implemented' blocks with functional logic based on the design sketch.

ps-architect