graph-mail
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFECOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads external, untrusted data from Microsoft 365 email bodies and calendar events which are subsequently processed by the agent.
- Ingestion points: The skill utilizes
graph-mail mail read,graph-mail mail search, andgraph-mail cal getto retrieve message and event content. - Boundary markers: No delimiters or safety instructions are provided to distinguish between retrieved content and system instructions.
- Capability inventory: While the
graph-mailtool itself is read-only, the agent may possess other write-capable skills (such asoutlook-cli) that could be targeted by an injection attack. - Sanitization: The skill does not perform any sanitization or filtering of the retrieved content before it enters the agent's context.
- [CREDENTIALS_UNSAFE]: The instructions describe storing sensitive authentication tokens in the local file
~/.config/graph-mail/tokens.jsonwhen specific environment variables or device-code flows are used. - [COMMAND_EXECUTION]: The skill provides examples of executing
python3 -cto parse and format JSON data retrieved from temporary files.
Audit Metadata