do-cleanup-swift
Pass
Audited by Gen Agent Trust Hub on Sep 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes
git statusandgit diffcommands to identify and analyze session changes. These operations are restricted to the local repository and are appropriate for the skill's intended purpose of code cleanup. - [INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection because it processes external Swift code via
gitcommand outputs, which could allow instructions hidden in code to influence the sub-agents. This risk is assessed as safe due to the skill's mandatory user-approval workflow. - Ingestion points: Modified Swift files identified via
git statusandgit diffinSKILL.md. - Boundary markers: The sub-agent prompt template uses clear framing ('You are reviewing Swift code... exclusively through your lens') to scope the analysis.
- Capability inventory: The skill uses
gitfor status/diff and the 'Skill tool' to load specialized sub-agents. It proposes edits but does not apply them automatically. - Sanitization: The skill relies on manual human review and approval of all proposed changes, preventing automated execution of injected instructions.
Audit Metadata