jmb-sync
Pass
Audited by Gen Agent Trust Hub on Nov 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local shell script (
pending-review.sh) located in a sibling directory using thebashcommand to evaluate the status of pull requests. - Evidence:
bash <this skill's base directory>/../jmb-review-pr/pending-review.sh <pr> >/dev/nullinSKILL.md. - [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources, including GitHub PR branch names, ticket numbers, and configuration fields from
docs/agents/flow.md, and interpolates them directly into a prompt for a new subagent. - Ingestion points:
docs/agents/flow.md,gh pr listoutput, and branch names fromgit worktree list. - Boundary markers: None. The data is concatenated directly into the subagent's instructions.
- Capability inventory: The skill can launch subagents, execute shell commands, and modify the file system via worktree operations.
- Sanitization: No sanitization or validation of the input fields is performed before prompt construction, which could allow a malicious branch name or documentation entry to influence subagent behavior.
Audit Metadata