deslop-duplication-audit
Warn
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and install the 'deslop' CLI tool from a third-party GitHub repository (github.com/Nimblesite/Deslop) using cargo or mise. These sources are not within the recognized trusted vendor scope.
- [COMMAND_EXECUTION]: The helper script 'deslop_report.dart' executes shell commands provided via the '--diff-cmd' flag. The implementation in 'lib/src/runner.dart' uses 'Process.run' with 'runInShell: true', which allows for shell command chaining and potential command injection if the input string is manipulated.
- [REMOTE_CODE_EXECUTION]: The skill relies on executing the 'deslop' binary, which is an external executable downloaded at runtime. Running unverified third-party binaries is a high-risk activity.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted code and diffs from external repositories to generate reports for the agent. This presents an attack surface where maliciously crafted source code or diffs could attempt to influence the agent's behavior through the generated Markdown summary.
- Ingestion points: Reads files and diffs from the target repository ('targetDir', '--diff-file', '--diff-cmd').
- Boundary markers: None implemented in the report generation to delimit code content from instructions.
- Capability inventory: File system writes (refactoring), shell execution ('Process.run'), and network operations (cargo/mise/git).
- Sanitization: Paths are normalized, but code content is summarized without escaping or strict validation before being presented to the agent.
Audit Metadata