sensible-contribution-flow
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill includes an optional mechanism to fetch a peer skill (
intentional-dependency-choice) from the author's repository on GitHub. This process is guarded by mandatory user confirmation, utilizes immutable commit hashes for integrity, and validates the fetched content's structure before use. - [COMMAND_EXECUTION]: The workflow involves executing Git and package manager commands to prepare pull requests and commits. The skill defines extensive safety boundaries, such as suppressing package lifecycle scripts with
--ignore-scripts, using--force-with-leasefor push operations to prevent overwriting remote work, and strictly requiring human-only checkpoints for sensitive actions. - [INDIRECT_PROMPT_INJECTION]: The instructions explicitly define a security posture for handling untrusted data, such as tool outputs or external repository content. It requires the agent to delimit untrusted material and forbids it from granting authority or expanding the task scope, directly mitigating prompt injection risks.
- [CREDENTIALS_SAFE]: The skill contains a dedicated policy for secrets and authentication. It prohibits the agent from directly handling, exposing, or requesting credentials, tokens, or private keys, ensuring all authentication management remains under the user's direct control.
Audit Metadata