omh-workflow-learning
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses framework-native CLI tools (
omhandhermes) to classify learning signals and audit skill modifications. These operations are consistent with the skill's stated purpose as a workflow tracker and are performed within the local environment. - [INDIRECT_PROMPT_INJECTION]: The skill identifies a vulnerability surface by processing runtime "lessons" and ledger entries to generate workflow traces.
- Ingestion points: Reads audit data from
skills/.curator_ledger.jsonland accepts user-provided lesson summaries as input for routing. - Boundary markers: Implements
prepared-vs-observedandnot-evidenceboundaries to separate AI guidance from actual runtime results. - Capability inventory: Uses
omhandhermesCLI commands to manage metadata-only records and review queues across the workflow. - Sanitization: Emphasizes the use of metadata-only records and manual review checkpoints instead of automatic store mutations.
Audit Metadata