collaborative-writing

Warn

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill interacts with the local file system to read 'lexicon' files and save drafts using paths constructed from user input.
  • Evidence: The skill uses a variable '[job-slug]' provided by the user to build paths such as '/career-applications/[job-slug]/01-job-analysis.md' and '/career-applications/[job-slug]/05-cover-letter-draft.md'.
  • Risk: Without sanitization, the '[job-slug]' input could be manipulated (e.g., using '../../') to perform directory traversal, allowing the agent to read or overwrite sensitive files elsewhere in the user's home directory.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the lack of boundary markers for ingested external data.
  • Ingestion points: The agent collects 'rough writing' and 'bullet points' directly from the user and reads data from files located in '/lexicons_llm/' and '/career-applications/'.
  • Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore' commands to isolate user-provided content from the agent's core instructions.
  • Capability inventory: The agent has permissions to read and write files on the host file system.
  • Sanitization: Absent. There is no evidence of validation or filtering for the text provided by the user or read from local files before it is processed by the model.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 1, 2026, 01:07 AM