conductor-dev

Pass

Audited by Gen Agent Trust Hub on Oct 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
  • [COMMAND_EXECUTION]: The skill provides the agent with the ability to execute shell commands through scripts like compact_command.py, incremental_verify.py, and change_impact.py. These tools use the subprocess module to interact with the local environment. Analysis confirms these are intended for legitimate development tasks, such as interacting with git and the HarmonyOS devecocli toolchain. \n- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes external data, including user-defined product briefs (product.md) and documentation snippets retrieved via devecocli_docs_bridge.py. \n
  • Ingestion points: External documentation fetched by devecocli_docs_bridge.py and project-specific files like product.md and TASKS.md (referenced in SKILL.md and references/context-policy.md).\n
  • Boundary markers: The skill employs a documentation bridge (devecocli_docs_bridge.py) that compacts results and flags non-English content for translation rather than passing raw documentation directly into the context.\n
  • Capability inventory: The skill has significant capabilities, including arbitrary command execution (compact_command.py), file system modification (state_update.py), and project building/deployment (incremental_verify.py).\n
  • Sanitization: Documentation is truncated and filtered for relevance by the bridge script. The system relies on a rigid, phase-based state machine defined in SESSION_STATE.json and validated by validate_state.py to constrain agent actions.\n- [DYNAMIC_EXECUTION]: The script clean_ai_workflow.py utilizes the __import__ function for dynamic module loading. However, this is used strictly for the standard json library and does not involve loading untrusted or external modules at runtime. \n- [PRIVILEGE_ESCALATION]: The script incremental_verify.py includes logic to execute PowerShell scripts using the -ExecutionPolicy Bypass flag. This is used as a compatibility layer to ensure that local platform tools (such as .ps1 wrappers for DevEco Studio CLI) can be executed correctly across different development environments.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 3, 2026, 09:48 AM