write-outreach-draft-sequence
Pass
Audited by Gen Agent Trust Hub on Oct 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONNO_CODE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the prompt and external web sources, which could contain malicious instructions.
- Ingestion points: The skill accepts inputs including brand data, segment angles, and "rows" containing signal and hook URLs in SKILL.md.
- Boundary markers: No explicit delimiters or instructions to ignore embedded commands within the input data or web content are present.
- Capability inventory: The skill is permitted to perform up to 3 web reads to verify the content of provided links in SKILL.md.
- Sanitization: No sanitization, validation, or escaping of the content retrieved from external URLs is defined.
- [PROMPT_INJECTION]: The skill contains instructions that explicitly restrict or override the agent's typical operational capabilities ("don't load any skill, don't read playbook files, don't start other agents, make no GooseWorks calls"). While appearing to be workflow constraints for this specific task, these are patterns used to manipulate agent behavior.
- [NO_CODE]: No executable scripts, binaries, or configuration-driven code were found; the skill relies entirely on natural language instructions.
Audit Metadata