tsg-flow-implementer

Pass

Audited by Gen Agent Trust Hub on Nov 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted instructions from project files such as tasks.md and TechSpecs. These files define the 'gate' commands that the agent executes in the shell. An attacker who can modify these files could execute arbitrary code on the system. * Ingestion points: tasks.md, flow-state.json, and various specification files (TechSpecs, ADRs). * Boundary markers: None. The skill treats the content of the task file as a direct source of truth for execution. * Capability inventory: Execution of arbitrary shell commands (gate_command), file modifications, and browser automation via Playwright. * Sanitization: No sanitization is performed on the commands read from external files; the instructions explicitly state "Não altere o comando" (Do not change the command).
  • [COMMAND_EXECUTION]: The skill's core functionality relies on executing shell commands provided in the task metadata. There are no constraints or sandboxing mentioned for these commands, which could perform sensitive operations or network exfiltration if the task definition is compromised.
  • [NO_CODE]: The skill does not bundle any executable scripts, relying instead on environment tools and markdown-based instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 17, 2026, 10:29 PM