cli-anything
Pass
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The sections titled 'Self-Evolving Skill' and 'Post-Execution Reflection' instruct the agent to modify the SKILL.md file directly based on its performance or observed errors. This creates a self-modification loop where the agent is encouraged to rewrite its own instructional logic.
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and analyze external software repositories (via GitHub URLs or local paths) to generate CLI code.
- Ingestion points: External codebase analysis via the
/cli-anything <path-or-url>command. - Boundary markers: No specific delimiters or instructions to ignore embedded directives within the analyzed source code are provided.
- Capability inventory: The skill utilizes the 'Read' tool and involves shell-based operations like 'pip install' and 'pytest'.
- Sanitization: No explicit sanitization or validation of the ingested source code is mentioned before the generation process.
- [EXTERNAL_DOWNLOADS]: The installation guide directs the user or agent to clone the CLI-Anything repository from 'github.com/HKUDS/CLI-Anything'. While this is a research-oriented repository, it involves downloading and installing external code and plugins.
- [COMMAND_EXECUTION]: The skill provides instructions for the agent to execute various system-level commands, including 'git clone', 'pip install', and running generated test suites using 'pytest'.
Audit Metadata