agent-orchestration-handoff
Pass
Audited by Gen Agent Trust Hub on Aug 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external project files, which could lead to indirect prompt injection if those files contain malicious instructions intended to influence subsequent agent sessions.
- Ingestion points: The skill instructs the agent to read project configuration files (such as CLAUDE.md or AGENTS.md) and previous handoff files from the filesystem.
- Boundary markers: The procedure requires the agent to output the handoff within a fenced code block to delineate it from other conversation content.
- Capability inventory: The skill involves reading multiple workspace files and writing summary content to a new file in the project root or temporary directory.
- Sanitization: The instructions explicitly mandate the redaction of API keys, tokens, and PII, and tell the agent to treat claims as context to verify rather than facts to trust blindly.
Audit Metadata