The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to handle "untrusted public content" and "audit every claim," creating an attack surface for malicious instructions embedded in external data.
Ingestion points: User-provided tasks and external data processed during "triage" or "adversarial verification" workflows (referenced in SKILL.md).
Boundary markers: The skill encourages the use of rubrics and focused goals for subagents to mitigate drift, though strict delimiters for external content are not explicitly detailed in the skeletons.
Capability inventory: The generated scripts utilize the agent() primitive which can spawn subagents with access to the worktree and potential file mutation capabilities.
Sanitization: The skill explicitly recommends a "quarantine" pattern where agents reading untrusted content are barred from performing high-privilege actions.
[DYNAMIC_EXECUTION]: The core functionality of the skill involves authoring and launching JavaScript-based orchestration scripts.
Evidence: references/pattern-catalog.md provides multiple code skeletons that use agent(), parallel(), and pipeline() primitives to execute logic.
Context: This dynamic execution is the primary intended purpose of the skill for task orchestration and management within the platform's workflow tool environment.

aep-workflow