wiki-rebuild

Warn

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell commands via a CLI tool, falling back to qmd if the QMD_CLI environment variable is unset. It performs operations such as update, embed, ls, and get. This represents a risk if the environment variable or paths are manipulated.
  • [INDIRECT_PROMPT_INJECTION]: The skill employs a 'Config Resolution Protocol' that searches parent directories for .env files. This creates a significant vulnerability surface.
  • Ingestion points: The agent reads .env files from arbitrary parent directories based on the current working directory.
  • Boundary markers: The skill relies on user confirmation for destructive actions but lacks technical validation for the configuration values loaded from external files.
  • Capability inventory: The skill can delete all content in category directories (concepts/, entities/, etc.), remove the .manifest.json file, and execute shell commands.
  • Sanitization: There is no evidence of path validation or shell argument escaping for the QMD_CLI command or the $QMD_WIKI_COLLECTION variable.
  • [COMMAND_EXECUTION]: The skill includes instructions to 'Clear live wiki,' which involves removing all content from multiple directories. While this is an intended feature, the lack of path sanitization makes it a high-risk capability that could be misdirected to delete system files if OBSIDIAN_VAULT_PATH is poisoned.
  • [DATA_EXFILTRATION]: The skill explicitly instructs the agent to read .env files and the .manifest.json file. While intended for configuration, these files often contain sensitive metadata or secrets, and the automated resolution process increases the risk of unauthorized data access.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 6, 2026, 01:27 AM