The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a Python utility script, scripts/check_vault.py, which executes git diff via the subprocess.run method. The arguments for these calls are derived from metadata headers such as Fingerprint and Monitored. Although the script uses a list-based argument format to mitigate shell injection, malicious values in these user-defined fields could still potentially be used to pass unexpected flags to the Git command.
[INDIRECT_PROMPT_INJECTION]: The skill instructs agents to process and compile data from Markdown files in the raw/ and wiki/ directories. These files could serve as an attack surface if they contain untrusted content designed to influence agent behavior. The vault's grounding logic provides a mitigation by requiring exact verbatim matches for numbers and quotes against immutable source files.
[EXTERNAL_DOWNLOADS]: The documentation references an external GitHub repository, github.com/PALAN-K/llm-wiki-loop, as a reference implementation for the vault architecture and workflow.

grounded-vault