write-project-readme
Warn
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/check_doc_commands.pyparses and executes shell commands from Markdown code blocks (sh, bash, shell, console). This functionality allows for the execution of arbitrary code found within a repository. - Evidence: Use of
subprocess.run(["sh", "-c", check.command], ...)inscripts/check_doc_commands.py. - [DYNAMIC_EXECUTION]: The skill performs runtime execution of shell commands dynamically extracted from Markdown files.
- [INDIRECT_PROMPT_INJECTION]: The skill processes content from repository files (Markdown, manifests) that may be controlled by an attacker.
- Ingestion points: Markdown files (e.g.,
README.md) and repository manifests. - Boundary markers: The skill relies on an informal
<!-- doc-check: skip -->convention to avoid executing dangerous commands, which is not a robust security boundary. - Capability inventory: Ability to execute shell commands, read/write files, and traverse the repository structure.
- Sanitization: No sanitization is performed on command strings extracted from Markdown blocks before execution.
Audit Metadata