The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents an auto-routing mechanism that processes user requests to activate specific agents and skills based on keywords.
Ingestion points: User requests are processed by the routing logic and the skill-activation-prompt.sh lifecycle hook.
Boundary markers: The architectural overview does not specify the use of delimiters or instructions to ignore embedded prompts during routing.
Capability inventory: The documented system can invoke sub-agents via the Task tool, read local audit logs, and execute monitoring scripts.
Sanitization: Input validation logic is not explicitly detailed in the meta-level architectural description.
[EXTERNAL_DOWNLOADS]: The documentation references an external GitHub repository for architectural patterns (github.com/frankxai/knowledge-work-plugins). This repository belongs to the skill's author and is cited as the source for design principles like connector agnosticism and progressive disclosure.

acos-meta