writing-fragments

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill performs local file system operations, including reading and appending data to a path provided by the user. This functionality is intended for note-taking but represents a data exposure surface if the agent is directed toward sensitive files.
  • Evidence: 'ask once where to save the document', 'always re-read it before writing so their edits are preserved', and 'Before every write: re-read the file from disk.'
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is explicitly instructed to ingest and process data from a file that could contain malicious instructions.
  • Ingestion points: Markdown file at a user-defined path (referenced in SKILL.md).
  • Boundary markers: Absent. There are no instructions to use delimiters or to treat the file content as inert data rather than instructions.
  • Capability inventory: The skill has file read and file append capabilities across its operating logic.
  • Sanitization: Absent. There is no evidence of validation or escaping for the content read from the disk before the agent processes it.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 06:43 AM
Security Audit — agent-trust-hub — writing-fragments