project-harness-init
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from a YAML block (
BOOTSTRAP_BRIEF) and uses it to populate multiple project files. - Ingestion points: The agent parses the
# BOOTSTRAP_BRIEFYAML block from the chat context (described inSKILL.mdFase 0). - Boundary markers: While it looks for a specific block header, it does not utilize delimiters or specific instructions to ignore malicious content embedded within the brief's values.
- Capability inventory: The skill performs extensive file-write operations across the workspace, creating documentation and configuration files like
feature_list.json,AGENTS.md, andCHECKPOINTS.md(SKILL.mdFase 2). - Sanitization: The instructions do not specify any validation or sanitization for the inputs (such as
stack.verify_commandorobjective) before they are written to the disk, creating a risk that malicious instructions could be persisted in the project's development context.
Audit Metadata