writing-fragments
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFEPROMPT_INJECTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core operational logic. It is instructed to read the existing contents of a user-specified markdown file before every write operation to preserve manual edits. This create a feedback loop where instructions embedded within a fragment (intentionally or otherwise) could be interpreted by the agent as system-level guidance during the next interaction cycle.
- Ingestion points: User-provided text fragments and the contents of the target markdown file.
- Boundary markers: There are no delimiters or instructions to treat the read file content as untrusted data.
- Capability inventory: File system read and write access.
- Sanitization: No validation or filtering of the file content or user fragments is implemented.
- [NO_CODE]: The skill contains only natural language instructions in SKILL.md and does not include any executable scripts, binaries, or complex configuration files. Its safety is largely dependent on the underlying agent's handling of file system tools.
Audit Metadata