agent-swarm-memory-manager
Pass
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION] (LOW): The skill is highly susceptible to indirect prompt injection due to its core function of managing shared memory across a 'hive mind'. If any agent in the swarm ingests malicious instructions from an external source and persists them to the shared coordination namespace, this skill will propagate those instructions to other agents without validation.
- Ingestion points: Data is retrieved from the
coordinationnamespace in thebatchReadandatomicWritepseudocode blocks withinSKILL.md. - Boundary markers: None. There are no instructions or delimiters used to separate data from instructions when reading from the memory tool.
- Capability inventory: The skill utilizes the
mcp__claude-flow__memory_usagetool to store, retrieve, and broadcast data across the agent network. - Sanitization: None. Data is handled as raw JSON and re-broadcasted or cached without escaping or verification.
- [TIME_DELAYED_ATTACKS] (SAFE): While the skill contains instructions to perform actions at specific time intervals (e.g., 'EVERY 60 SECONDS' and 'every 30 seconds'), these triggers are used solely for writing performance metrics and synchronizing state to the internal memory tool. These do not gate any high-risk operations like network exfiltration or command execution.
Audit Metadata