balance-test
Warn
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: MEDIUMPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PERSISTENCE]: The skill implements a self-evolution telemetry system that writes execution metadata to
~/.claude/projects/. This allows the skill to maintain a history of its operations and outcomes across different sessions by storing data in the user's home directory. - [DATA_EXPOSURE]: The skill scans for project memory directories within
~/.claude/projects/. This involves searching the file system outside of the immediate workspace to find and read telemetry files, exposing internal agent metadata. - [COMMAND_EXECUTION]: During the 'Self-Healing Validation' phase, the agent is instructed to run the project's test suite and execute build/compile commands. These operations involve running arbitrary code or scripts defined within the analyzed repository.
- [INDIRECT_PROMPT_INJECTION]: The skill acts as an autonomous agent that ingests large amounts of untrusted data from the project codebase (stats, economy tables, RNG logic).
- Ingestion points: Codebase files (scripts, JSON/YAML configs, data tables) are parsed in Phase 1.
- Boundary markers: Absent. There are no instructions to the agent to disregard instructions that might be embedded within the game data files it extracts.
- Capability inventory: The skill can execute shell commands for building, compiling, and running tests as defined in the 'Self-Healing Validation' section.
- Sanitization: None. Extracted stats are used directly in mathematical simulations and reports.
Audit Metadata