oma-docs
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the local codebase to generate documentation patches via the host LLM, creating a potential injection vector.
- Ingestion points: The skill reads documentation files (
docs/**/*.md) and ingestsgit diffoutput to identify changes. - Boundary markers: It respects explicit escape hatches such as
<!-- oma-docs:ignore-start -->/<!-- oma-docs:ignore-end -->and YAML frontmatteroma-docs: skipto exclude specific regions from processing. - Capability inventory: The skill has the ability to write to the file system by applying patches via
git applyor direct file writes, and it executes shell commands for diagnostics. - Sanitization: A specific guardrail redacts secret-bearing files (e.g.,
.env*,*.pem,*.key,id_rsa*) and gitignored files from being included in the diff intake for the LLM. - [COMMAND_EXECUTION]: The skill relies on executing system commands to perform file analysis and state management.
- Execution: It spawns subprocesses for
git diff,git apply,which, and the external link-checking toollychee. - User Control: In
syncmode, all file modifications are interactive, requiring the user to explicitly confirm each patch with a[y]response before any changes are committed to the disk.
Audit Metadata