The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill reads from external specification files and injects their content directly into tasks for other agents, which is an attack surface for indirect prompt injection.
Ingestion points: The workflow (Step W01) ingests content from SPEC files located at llmwiki/wiki/sources/draft/DDMMYY-<tên>.md.
Boundary markers: The skill does not provide clear delimiters or instructions to treat the ingested SPEC content as untrusted data; it explicitly requires verbatim (nguyên văn) injection into downstream tasks.
Capability inventory: The skill utilizes the orca tool for task-create and dispatch --inject operations, which can execute commands or provide instructions to other agents.
Sanitization: There is no evidence of sanitization, filtering, or validation performed on the SPEC content before it is injected into the task dispatcher.
[COMMAND_EXECUTION]: The skill interacts with the orca orchestration CLI to automate task management.
Evidence: The skill workflow and action steps (W05, W08) detail the execution of orca orchestration task-create and orca dispatch --inject to manage agent tasks.

plan