This workspace contains the knowledge base, Excel test-case template, and Burr workflow used to generate CANoe automated test project artifacts.
Current architecture details are maintained in
docs/ARCHITECTURE_CURRENT.md. The older reports in docs/ remain useful as
historical evaluations, but the current source of truth is the code plus the
architecture document.
See CHANGELOG.md for the latest workflow changes.
CANoe_Cfg_AutoGene/
CANoeCANalyzer.chm
ci/
docs/
knowledge_base/
templates/
workflows/
generated_projects/ # local regenerated outputs, git-ignored
Required to run the default workflow and knowledge-base checks:
- Windows with Python 3.9+.
burrfor the workflow state machine.rank_bm25,jieba, andlxmlfor the default model-free knowledge-base retriever.- Git LFS when cloning the repository, because
CANoeCANalyzer.chmandknowledge_base/artifacts/bm25.pklare LFS-managed files.
Install the required Python packages with the same Python used to run the workflow:
python -m pip install -r requirements.txtOptional dependencies:
- Node.js is only needed when rebuilding
templates/canoe_test_case_template/CANoe自动化测试用例模板.xlsx. - Vector CANoe plus a project-specific adapter is only needed for real CFG loading or CAPL compilation. With
--canoe-validation-mode disabled, the workflow does not require CANoe and reports external validation as skipped. knowledge_base\requirements-optional.txtis only for explicit vector/rerank/MCP experiments; the default path does not need embedding models.
-
knowledge_base/
Vector CANoe/CAPL knowledge base, agent-facing indexes, retrieval tools, and workflow knowledge contracts. The workflow uses scoped profiles so CANoe CFG generation retrievescanoe_configfacts, while CAPL authoring retrievescapl_authoringfacts. -
templates/canoe_test_case_template/
Maintained Excel test-case template, shared field mapping, and generator script. -
workflows/canoe_auto_generation_burr/
Apache Burr workflow implementation for parsing the Excel template and producing structured test cases, source models, coverage reports, CANoe configuration plans/scripts, CAPL test module source, evaluation reports, correction workbooks, HTML run reports, repair plans, and final manifests. Supporting modules split out source parsing, CAPL lexer linting, report generation, run retention, canonical field IDs, and external CAPL agent adapters. -
ci/
Local validation entry points.python ci/run_checks.pyvalidates the Burr profile, workflow knowledge references, Python syntax, and retrieval profile isolation. -
docs/
Workspace-level architecture and evaluation reports.
generated_projects/
Workflow run outputs. These are reproducible artifacts, are ignored by Git, and can be regenerated from the template and workflow.
cd E:\Canoe_Gene\CANoe_Cfg_AutoGene
python -m workflows.canoe_auto_generation_burr.canoe_workflow `
--excel .\templates\canoe_test_case_template\CANoe自动化测试用例模板.xlsx `
--out .\generated_projects\EQ07_workflow_test `
--capl-authoring-mode llm `
--canoe-validation-mode disabledEach run is isolated under generated_projects/<project>/runs/<run_id>. The
base output directory keeps latest_run_manifest.json as a pointer to the most
recent run.
Main generated artifacts include:
structured_test_cases.json: parsed Excel test-case IR.source_models.json: lightweight DBC/A2L/CDD/CFG/DLL source model summary.evidence_bundle.json: CAPL API evidence retrieved through the active KB profile.coverage_report.json: requirement, feature, domain, source, and evidence coverage summary.<project>.cfg: copied base CFG or CANoe-generated CFG when automation runs.<project>.cfg.todo.json: configuration plan when no base/generated CFG is available.<project>.cfg.plan.json: CANoe configuration generation plan.<project>.cfg.generate.ps1: CANoe COM Automation script for CFG SaveAs.<project>_TestModule.can: generated CAPL test module.capl_script_plan.json: CAPL authoring plan, evidence refs, coverage summary, adapter gaps, and LLM agent result.run_report.html: HTML summary for human review.test_case_corrections.xlsx/.md: correction package when the run is blocked.
# Rebuild the Excel template
node .\templates\canoe_test_case_template\build_template.mjs
# Validate workflow knowledge references
python .\knowledge_base\workflow_kb\validate_workflow_kb.py
# Validate workflow_profile.json against the executable Burr graph
python .\workflows\canoe_auto_generation_burr\validate_workflow_profile.py
# Compile workflow Python files
python -m compileall .\workflows\canoe_auto_generation_burr
# Run the project check bundle
python .\ci\run_checks.pyUseful workflow flags:
--strict-source-validation: missing or unresolved DBC/A2L/CDD semantics become blocking errors and producerepair_plan.json.--canoe-validation-mode disabled|manual|automated: controls the Vector CANoe external validation adapter. Disabled/manual never claim CANoe compile success.--capl-authoring-mode llm: requires the external LLM agent path for CAPL generation. Planner/Fixer commands can be split with environment overrides.--capl-compile-max-attempts N: maximum Planner/Fixer compile-repair attempts insidegenerate_capl_script.--quality-gate exploration|release: release mode rejects unboundCanoeGene_*adapter stubs.--tracking: enables Burr local tracking for action-level debugging.
Run from the repository root:
py -3.9 ci\run_checks.pyLast verified locally on 2026-06-22 (Python 3.9, burr 0.40.2):
validate_workflow_profile.py→ 14 actions / 19 transitions, profile 与代码一致。validate_workflow_kb.py→ 30 references, 0 missing, 0 invalid JSON。py_compile→ 20 个.py文件全部通过。test_retrieval_profiles.py→ 2 个用例通过。python -m unittest discover -s workflows/canoe_auto_generation_burr -p "test_*.py"→ 26/26 通过。
burr==0.40.2当前不支持 Python 3.14;推荐使用py -3.9。
The CAPL generation step can call an external KB-indexed LLM agent without
giving it access to the whole repository or full knowledge base. The repository
ships a Claude Code adapter at
workflows/canoe_auto_generation_burr/agents/claude_code_capl_agent.py:
$env:CANOE_GENE_CAPL_AGENT_COMMAND = 'python -m workflows.canoe_auto_generation_burr.agents.claude_code_capl_agent "{payload}" "{response}"'
$env:CANOE_GENE_CLAUDE_CODE_COMMAND = 'claude.cmd'generate_capl_script uses two roles. Planner runs once and returns immutable
golden_ir; Fixer runs in the compile loop and receives only golden_ir, the
current complete CAPL source, and the latest compile error log. Optional
CANOE_GENE_CAPL_PLANNER_COMMAND and CANOE_GENE_CAPL_FIXER_COMMAND override
the shared command per role.
Planner response:
{
"golden_ir": {
"project": {},
"channels": [],
"cases": [],
"capl_api_plan": [],
"assertions": [],
"branching": [],
"evidence_refs": []
}
}Fixer response:
{
"capl_source": "complete .can source",
"capl_script_plan": {
"assumptions": [],
"adapter_gaps": [],
"used_evidence_refs": [],
"cases": []
}
}If the command is not configured, the workflow records the unavailable agent in
capl_script_plan.json and emits an llm_failed CAPL artifact for evaluation.
The Claude Code adapter accepts these optional overrides:
CANOE_GENE_CLAUDE_CODE_MODELCANOE_GENE_CLAUDE_CODE_EFFORTCANOE_GENE_CLAUDE_CODE_TIMEOUT_SECONDSCANOE_GENE_CLAUDE_CODE_EXTRA_ARGS
To feed CANoe compile diagnostics back to Fixer, set automated validation. The
workflow writes and runs an internal PowerShell COM script that opens the
generated CFG when available, attaches the CAPL file, tries
Configuration.RunCompilation, Configuration.Compile, then
$canoe.CAPL.Compile(), and reads diagnostics from the available
compile-result COM object:
python -m workflows.canoe_auto_generation_burr.canoe_workflow `
--excel .\templates\canoe_test_case_template\CANoe自动化测试用例模板.xlsx `
--out .\generated_projects\EQ07_workflow_test `
--canoe-validation-mode automated `
--capl-compile-max-attempts 5CANOE_GENE_CAPL_COMPILE_COMMAND remains available as an override hook. When
set, it must write JSON to {report} with status: "pass" or status: "failed". Only the latest failed compile report is passed into the next Fixer
prompt.
The workflow does not hand-write Vector's private .cfg serialization format.
It discovers a base CFG when available, prepares a .cfg.plan.json, and emits a
PowerShell COM Automation script that opens CANoe, mounts supported DBC/CDD/A2L
and DLL inputs, then calls CANoe Configuration.SaveAs.
With --canoe-validation-mode disabled, the script is generated but not run.
Use --canoe-validation-mode automated only on a machine where CANoe COM
Automation is available and the project can be safely opened.
Excel dropdown values and Burr workflow validation enums share one source:
templates/canoe_test_case_template/template_field_mapping.json