The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted data from external URLs or files into a knowledge base. This creates a surface for indirect prompt injection where instructions hidden in sources could influence the agent during the compilation or querying phases. The skill mitigates this risk through detailed 'distillation' and 'filtering' procedures described in references/ingest.md and references/distilled-ingest.md, which instruct the agent to remove filler, strip PII, and reorganize content rather than copying it verbatim. Per analysis rules, this vulnerability surface is documented but assigned a low severity as it is inherent to the skill's purpose and addressed by internal guidelines.
[DATA_EXFILTRATION]: The skill includes a robust security feature called the 'local-content leak guard' (implemented in scripts/lint_wiki.py and references/lint.md). This mechanism scans the wiki for any links that point to a private local/ directory, preventing the accidental exposure or exfiltration of uncommitted private notes into the shared git repository.
[COMMAND_EXECUTION]: The skill utilizes two Python scripts (scripts/lint_wiki.py and scripts/lint_mermaid.py) to perform structural validation. These are executed via the uv tool. Analysis of these scripts shows they are written using only the Python standard library and contain no network operations, privilege escalation, or malicious file system access; they are strictly read-only tools for maintaining wiki health.

llm-wiki