the-sequence-doctor
Pass
Audited by Gen Agent Trust Hub on Aug 14, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill processes untrusted content such as email replies, transcripts, and web pages, which are potential vectors for indirect prompt injection.
- Ingestion points:
SKILL.mdandreferences/agent-security.mdidentify fetched pages, pasted exports, transcripts, and inbound replies as untrusted content sources. - Boundary markers: Robust defensive instructions are present, explicitly directing the agent to treat content as data, never instructions, and to quote rather than summarize injection attempts to prevent propagation.
- Capability inventory: The agent performs text-based analysis and rewrites. It has no capabilities for unauthorized network operations or file-system modifications triggered by the analyzed data.
- Sanitization: The instructions require HTML-escaping for emitted markup and explicitly forbid following URLs found within retrieved content.
- [DATA_EXPOSURE]: The skill contains logic to detect and mitigate the accidental exposure of credentials found in user-provided data.
- Evidence:
SKILL.mdandreferences/agent-security.mdprovide clear rules for the agent to identify potential API keys or tokens in inputs and report them for rotation without reproducing or persisting the secrets. - [SAFE]: External references to
intempt.comand related tooling are consistent with the skill's stated authorship and functionality.
Audit Metadata