micropatch
Pass
Audited by Gen Agent Trust Hub on Sep 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill architecture requires the agent to read and follow instructions contained within external MicroPatch specification files (
micropatch.md) to apply customizations to a codebase. - Ingestion points: The agent ingests data from
micropatch.md,hints.diff, and implementation artifacts in afiles/directory as described inSKILL.mdandreferences/apply.md. - Boundary markers: The instructions do not define boundary markers or include "ignore embedded instructions" warnings for the agent when processing these external specification files.
- Capability inventory: The agent is directed to perform file system modifications and execute arbitrary verification commands (such as shell commands, tests, and API checks) as specified in the patch's "Definition of Done" section in
references/apply.md. - Sanitization: There is no mention of sanitizing or validating the implementation logic, code references, or verification commands provided by the external patch data.
Audit Metadata