statechart

Pass

Audited by Gen Agent Trust Hub on Nov 8, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches essential toolsets from trusted technology providers.
  • Downloads the TLA+ model checking engine (tla2tools.jar) from the official Tlaplus organization's GitHub releases.
  • Utilizes the mise manager to install the Java Development Kit (JDK) from well-known sources at runtime.
  • [COMMAND_EXECUTION]: Executes local utilities and tool binaries to process formal specifications.
  • Spawns java processes to run the SANY specification parser and the TLC model checker.
  • Executes system-level commands like rm and taskkill to manage temporary files and ensure clean process termination.
  • [DYNAMIC_EXECUTION]: Supports custom visualization modules via local code loading.
  • Imports and executes user-provided JavaScript or TypeScript modules (passed via the --view flag) to customize state labels and chart metadata.
  • [REMOTE_CODE_EXECUTION]: Runs a verified Java archive downloaded from a trusted source.
  • Executes the tla2tools.jar binary fetched from GitHub, which is a required dependency for formal system analysis.
  • [INDIRECT_PROMPT_INJECTION]: Processes formal specification data that is subsequently reflected in markdown reports.
  • Ingestion points: Model variable values and action names are parsed from user-supplied TLA+ specifications (SPEC.tla).
  • Boundary markers: The generated report.md includes values from the model without specific delimiters to isolate them from the surrounding context.
  • Capability inventory: The skill scripts have access to shell command execution and filesystem write operations.
  • Sanitization: Data extracted from the model is included in reports based on the user-provided view logic without extensive filtering, though the formal nature of TLA+ reduces the typical risk surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 8, 2026, 07:22 PM