The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted Vietnamese text from the user or files and passes it to a multi-agent pipeline (Diagnostic, Polisher, and Critic agents) for analysis. The instructions for these sub-agents do not include boundary markers or delimiters for the untrusted content. Combined with the skill's access to powerful tools such as Bash and WebFetch, this creates a potential surface where malicious instructions embedded in the input text could influence the agent's behavior.
Ingestion points: The primary Vietnamese text input passed as a string or file path to the orchestrator in SKILL.md.
Boundary markers: Absent. The input is passed directly to sub-agents without delimiters.
Capability inventory: The skill has access to Read, Grep, Glob, Bash, and WebFetch tools.
Sanitization: None detected. The text is evaluated for tone and register as provided.
[COMMAND_EXECUTION]: The skill includes a local TypeScript script (scripts/bootstrap-experience.ts) designed to be executed via bun to set up the context storage environment (.agents/experience/). While the script implements security best practices—such as using lstatSync to check for symbolic links to prevent path traversal—the requirement for the agent to resolve absolute paths and execute local setup scripts represents a powerful capability that requires monitoring.

vn-tone