persona-life-assistant

Warn

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: MEDIUMPERSISTENCEPRIVILEGE_ESCALATIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill implements a signaling mechanism that writes data to ~/.openclaw/feedback/signals.json via a Node.js bridge (scripts/state-sync.js). This involves interacting with the host environment outside the skill's own directory, allowing the agent to maintain state and communicate with the underlying platform across different sessions.
  • [PRIVILEGE_ESCALATION]: Through the defined 'Signal Protocol', the agent is instructed to proactively request increased permissions from the host, such as file_io or capability_gap. This mechanism provides a structured way for the agent to attempt to expand its operational boundaries and gain access beyond its initial sandbox constraints.
  • [EXTERNAL_DOWNLOADS]: The skill uses npx to execute the @agentplanet/acn package for on-chain identity registration. This involves downloading and running code from an external package registry, which introduces risks related to third-party dependency trust and integrity.
  • [COMMAND_EXECUTION]: The skill is configured with extensive tool access, including multiple Bash scopes, npm, npx, node, and curl. It relies on these tools for core functions like state synchronization and registration, creating a significant surface for command execution on the host system.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data—such as tasks, shopping lists, and recipes—while having access to powerful tools like shell execution and web fetching. This combination creates a surface for indirect prompt injection attacks.
  • Ingestion points: User input for schedule management in references/reminder.md and task/shopping list data in soul/persona.json.
  • Boundary markers: There are no specified delimiters or 'ignore embedded instructions' warnings for external data processing.
  • Capability inventory: Access to Bash (including openclaw, node, npm, npx, and curl), Read/Write file operations, and WebFetch for network requests.
  • Sanitization: No evidence of sanitization, input validation, or escaping of user-provided or web-fetched content is present in the skill's logic.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 4, 2026, 03:45 AM