improve-animations
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for indirect prompt injection as it is designed to ingest and analyze untrusted repository content. It implements the following mandatory security controls:
- Ingestion points: Source code files are read and processed during the 'Recon' and 'Audit' phases (SKILL.md).
- Boundary markers: Hard Rule 4 explicitly defines repository content as data rather than instructions, establishing a logical boundary for the model.
- Capability inventory: The primary advisor role is strictly read-only on source code; writes are restricted to a dedicated plans directory, and implementation is delegated to a scoped subagent.
- Sanitization: The agent is instructed to treat attempts at instruction overriding (such as 'ignore previous instructions') as findings to be reported rather than instructions to be followed.
- [COMMAND_EXECUTION]: The skill manages implementation through a delegation model using the
execute <plan>invocation. This dispatches a separate subagent to apply changes in an isolated worktree, ensuring the main advisor model retains a read-only posture relative to the primary source tree. - [SAFE]: The skill includes explicit prohibitions against high-risk activities. Hard Rule 2 forbids mutating operations including package installations, builds with side effects, and unauthorized commits, adhering to best practices for least-privilege auditing tools.
Audit Metadata