git-repo-digest
Fail
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill relies on a bundled C# script (
scripts/digest.cs) which is executed viadotnet run. This script has the capability to execute further system commands, includinggit cloneanddotnet test. Most critically, the validation phase involves the agent generating C# code blocks which are then written to disk and executed by the script usingdotnet test. This creates a path where malicious content in a processed repository could lead to the execution of arbitrary code on the host system. - [PROMPT_INJECTION]: The instructions in
SKILL.mdcontain several 'override' patterns. It explicitly tells the agent to 'Treat generated output as the source of truth' and to follow generated files (manifest.json,instructions.md, etc.) even if they 'disagree with this skill'. This instruction effectively allows external, dynamically generated content to bypass the agent's primary programming and safety guidelines. - [COMMAND_EXECUTION]: The
scripts/digest.csfile usesProcess.Startto executegitanddotnetCLI tools. While these are used for the skill's stated purpose of repository analysis, the parameters (URLs and project names) are derived from user input and repository metadata, posing a risk of command injection if not properly sanitized by the underlying OS. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from remote git repositories. Because the instructions demand the agent treat the results of the repository analysis (which include chunks of source code and READMEs) as authoritative instructions for its output, a repository containing hidden malicious instructions could manipulate the agent's behavior during the digest generation phase.
- [DYNAMIC_EXECUTION]: The workflow involves multiple stages of dynamic content generation and execution. The C# runner generates a
manifest.jsonandinstructions.mdwhich the agent is then required to follow to determine its next actions. Furthermore, theBasic usageexamples are dynamically placed into C# project templates and compiled/tested at runtime.
Recommendations
- AI detected serious security threats
Audit Metadata