The Agent Skills Directory

[PROMPT_INJECTION]: The file references/frame-worker-core.md contains explicit instructions for sub-agents to "supersede the skill catalog's own imperatives" and "do not open hyperframes/SKILL.md or hyperframes-core/SKILL.md". This is a form of instruction override designed to enforce narrow scope for child agents but matches patterns used to bypass systemic constraints.
[PROMPT_INJECTION]: The skill implements a multi-step orchestration chain where a main agent generates project files that subsequently serve as instructions for child agents (Indirect Prompt Injection).
Ingestion points: Sub-agents read tasks from STORYBOARD.md and BRIEF.md (via scripts/lib/frame-packets-core.mjs), which contain user-influenced content.
Boundary markers: The prompt construction logic and dispatch instructions do not implement delimiters or explicit warnings to ignore embedded instructions within user-provided storyboard blocks.
Capability inventory: Sub-agents have the capability to write HTML and JavaScript files to the compositions/frames/ directory, which are subsequently rendered or executed by the framework.
Sanitization: There is no evidence of sanitization or escaping of the content ingested from the markdown files before it is included in the sub-agent's prompt context.
[COMMAND_EXECUTION]: The skill defines workflows that execute shell commands using node for preference management (references/brief-contract.md) and npx hyperframes for project linting, validation, and rendering (SKILL.md, references/tailwind.md). These commands are part of the intended local developer workflow.
[EXTERNAL_DOWNLOADS]: The skill references external JavaScript libraries from well-known services: GSAP from cdn.jsdelivr.net and the Tailwind CSS browser runtime. These are used for legitimate project functionality but involve fetching remote code during the development or preview process.

hyperframes-core