adlc-orchestrator
Warn
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: MEDIUMPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Deceptive metadata detected in SKILL.json. The author field 'Anthropic Marketplace 3p' is inconsistent with the actual author '420company', potentially misleading users about the skill's origin and safety.
- [PROMPT_INJECTION]: Indirect prompt injection vulnerability. The skill gathers untrusted 'functional requirements' and interpolates them directly into agent tasks and shell commands without validation or boundary markers.
- Ingestion points: Requirements Gathering phase in SKILL.md.
- Boundary markers: Absent; requirements are passed directly into string templates like Task(adlc-author, 'Create agent from requirements: [requirements]').
- Capability inventory: Access to Bash and specialist agent delegation (Author, Engineer, QA).
- Sanitization: Absent; no validation or escaping of external requirements data before processing.
Audit Metadata