bmad-agent-dream-weaver
Pass
Audited by Gen Agent Trust Hub on Apr 13, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill ingests and processes user-provided dream narratives, which introduces a surface for indirect prompt injection. Maliciously crafted dream entries could theoretically influence the agent's interpretation or analysis logic in future sessions. \n
- Ingestion points: User input captured via dream-log.md and stored as markdown files in the journal/ directory. \n
- Boundary markers: The skill explicitly loads and enforces access-boundaries.md to restrict all file operations to the {project-root}/_bmad/memory/dream-weaver-sidecar/ directory. \n
- Capability inventory: The agent has permissions to read/write within the sidecar directory and execute the provided Python scripts for data analysis. \n
- Sanitization: Metadata within journal entries is processed using yaml.safe_load(), though the main dream narrative is processed as raw text for natural language interpretation.\n- [COMMAND_EXECUTION]: The skill executes local Python scripts (symbol_stats.py, recall_metrics.py, seed_tracker.py) for processing dream statistics and recall metrics. These scripts are part of the skill's core functionality, do not require network access, and operate exclusively on the data stored within the skill's authorized memory directory.
Audit Metadata