Skip to content
BUSMUST
中文Support

BMCLI AI integration

BMCLI ships one portable bmcli skill. Claude Code and Codex use the same SKILL.md and automotive workflow references; Codex additionally receives agents/openai.yaml for its own interface metadata.

Terminal window
bmcli ai setup
bmcli ai setup --target=codex
bmcli ai setup --target=claude
bmcli ai status --format=json

ai setup is offline and idempotent. It copies the bundled skill to $AGENTS_HOME/skills/bmcli (default ~/.agents/skills/bmcli) and $CLAUDE_CONFIG_DIR/skills/bmcli (default ~/.claude/skills/bmcli). It does not install BMCLI itself and does not modify project target/bench configuration.

The skill queries bmcli help <category> <action> --format=json at runtime. Command syntax and option defaults come from the installed binary, while the bundled references provide environment, development, verification/validation, and debugging methodology.

The skill uses two project files when initializing a test environment:

  • .bmcli/target.json: one portable target with named networks and addressed protocol nodes; it may be versioned.
  • .bmcli/bench.local.json: machine-local logical-network to analyzer-channel bindings, wiring values, and local paths; gitignore it.

Explicit user values override project configuration, which overrides local bench values, which override authorized discovery. Missing DBC data degrades to raw CAN IDs/payloads; missing safety-critical or state-changing values must not be guessed.

Source of truth: bmcli_reference.md in the BMCLI repository. Run bmcli help <category> <action> --format=json for the exact contract of the installed version.