ralph-loop-implement
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local CLI commands including
ralphandopencodeto perform repository operations, check state, and monitor progress. It uses shell scripting (loops withjq,wc, andtail) to track transcript activity and detect potential hangs or stalls during long-running tasks. - [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where untrusted data from GitHub issues drives automated implementation actions.
- Ingestion points: External data is ingested from the GitHub issue queue, specifically selecting issues with the
agent-readylabel (SKILL.md). - Boundary markers: There are no explicit instructions or delimiters defined to isolate the issue content or instruct the model to ignore embedded instructions within the issues.
- Capability inventory: The skill utilizes tools (
ralphandopencode) that are explicitly described as having the capability to change code, create commits, and close issues (SKILL.md). - Sanitization: The instructions do not specify any sanitization, filtering, or validation steps for the issue content before it is passed to the implementation engine.
Audit Metadata