yetong-fitness-xhs-profile
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to run a local Python script,
scripts/validate_profile.py, to verify the formatting and status of the generatedperson-dna.mdfile. The script uses standard libraries and performs basic regex checks on the file content. - [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted natural language data from user interviews to populate a persistent profile.
- Ingestion points: User responses during the progressive interview process outlined in
SKILL.mdandreferences/interview-guide.md. - Boundary markers: The instructions explicitly require the agent to separate original user quotes from its own summaries and use a strict Markdown/YAML schema to structure the data.
- Capability inventory: The agent performs local file writes (
person-dna.md) and executes a provided local validation script. - Sanitization: The skill implements a confirmation workflow where a profile must be explicitly verified by the user (transitioning from
drafttoactive) before being used by downstream skills, mitigating the risk of instructions embedded in the user's responses being blindly followed.
Audit Metadata