The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the llm-wiki sync command to finalize the research process. This is a legitimate functional requirement for maintaining the wiki vault's synchronization and state.
[DATA_EXFILTRATION]: The skill reads vault-specific configuration files (wiki-purpose.md, wiki-schema.md) and writes source materials to the local file system (sources/ directory). These actions are confined to the project workspace and align with the skill's research objectives.
[PROMPT_INJECTION]: The process of gathering 5–10 external sources introduces a surface for indirect prompt injection. The skill does not define boundary markers or sanitization procedures for this untrusted content, which could contain instructions intended to influence the agent. 1. Ingestion points: External URLs and files saved in the sources/ directory. 2. Boundary markers: Absent. 3. Capability inventory: File system writes across wiki pages and shell command execution (llm-wiki sync). 4. Sanitization: Absent.

research