leann
Pass
Audited by Gen Agent Trust Hub on Oct 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to perform local repository inspection and interact with theleannCLI for document indexing and retrieval. - [EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the
leannpackage viauv pip install, referencing official package registries for the tool's dependencies. - [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted text data to generate vector embeddings, presenting a potential surface for indirect prompt injection if source files contain adversarial instructions.
- Ingestion points: Local document files are read by the
leann buildcommand as specified inSKILL.mdand the upstream references. - Boundary markers: The skill relies on user-driven scoping of input directories; explicit prompt delimiters for document content are not provided in the skill instructions.
- Capability inventory: The skill employs
Bashfor operational commands and a local Python script for repository verification. - Sanitization: The
scripts/preflight.pyutility includes strict path validation and a blocklist (e.g.,.ssh,.aws,.env) to prevent accidental ingestion of sensitive credentials during the verify step. - [SAFE]: The included
preflight.pyscript implements robust security practices, including the use ofos.O_NOFOLLOWand directory-relative file descriptors to mitigate symbolic link attacks and race conditions during file verification.
Audit Metadata