dsh-plugin-dbc
用于 DeepSeek Harness 的 CAN DBC 编辑、验证、比较和 Excel 转换
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:wuwei6666/dbc-skill#6145d5c9e809484e86ff1225b413e983f5aefb25Plugin Dbc 以 dsh-plugin-dbc 发布，当前版本为 0.1.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
A structured command-line skill for creating, inspecting, updating, deleting, validating, and comparing CAN DBC databases, and for converting between DBC and Excel workbooks. It supports nodes, messages, signals, Classic CAN and CAN FD frame metadata, standard and extended multiplexing, JSON output, and deterministic node/message/signal comparison.
requirements.txt (cantools==40.7.1, openpyxl==3.1.5)From the repository root:
python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt
source .venv/bin/activate
Keep the repository layout intact when installing the skill: SKILL.md, references/, examples/, and scripts/ use relative paths. Run commands from the repository root.
Python 3.9+ is required because the pinned cantools==40.7.1 dependency requires it (openpyxl==3.1.5 requires Python 3.8+). Lowering this requirement would require changing and revalidating the DBC engine. Node.js 22+ is required for the npm wrapper.
After the package is published to npm:
# One-time setup: create an isolated Python environment and install dependencies.
npx --yes dsh-plugin-dbc setup
# Use from any directory; file paths remain relative to your working directory.
npx --yes dsh-plugin-dbc --json summary ./example.dbc
npx --yes dsh-plugin-dbc --json excel export ./example.dbc ./example.xlsx
npx --yes dsh-plugin-dbc --json excel import ./example.xlsx ./roundtrip.dbc
npx --yes dsh-plugin-dbc --json compare ./example.dbc ./roundtrip.dbc
# Install the native plugin into your Harness profile (tested CLI: 0.2.1-rc.2).
npx --yes @deepseek-ai/dsh web --help
npx --yes @deepseek-ai/dsh plugin --profile web add dsh-plugin-dbc
npx --yes @deepseek-ai/dsh web
The bundle registers the dbc tool with structured argument tokens and JSON results. The plugin runs Python on the host, so every invocation goes through Harness user approval; a surface without approval support denies the call. Use absolute paths in agent tool calls. The package does not contain an LLM or API key and does not send DBC files to an external conversion service. The agent's configured model still receives tool arguments and results.
setup stores its environment in ~/.dbc-skill/venv-v1 (override with DBC_SKILL_VENV). It runs only when explicitly requested, never as an npm install hook. To use an existing Python environment instead, set DBC_SKILL_PYTHON to its executable; both pinned dependencies must be installed. Python must already be installed; npx installs the JavaScript package, not Python.
For local development before publication:
npm ci
node bin/dbc-skill.js setup
npm test
npm pack
# Install the resulting tarball into Harness, without publishing.
npx --yes @deepseek-ai/dsh plugin --profile web add ./dsh-plugin-dbc-0.1.0.tgz
# Create and validate an empty database.
python3 scripts/dbc_crud.py --json init demo.dbc
python3 scripts/dbc_crud.py --json validate demo.dbc
# Add a node, a message, and a signal.
python3 scripts/dbc_crud.py --json node add demo.dbc ECU_A --comment "Body controller"
python3 scripts/dbc_crud.py --json message add demo.dbc 0x123 Status --length 8 --sender ECU_A
python3 scripts/dbc_crud.py --json signal add demo.dbc Status State --start 0 --length 8
# Inspect the result and validate again.
python3 scripts/dbc_crud.py --json signal show demo.dbc Status State
python3 scripts/dbc_crud.py --json summary demo.dbc
python3 scripts/dbc_crud.py --json validate demo.dbc
Message references accept a name or a decimal/0x hexadecimal frame ID. Put global options before the command, for example python3 scripts/dbc_crud.py --json --encoding utf-8 summary demo.dbc.
node, message, and signal add/update/delete operations save through cantools. An in-place save creates PATH.bak containing the original bytes by default; retain it while checking the result.--output OTHER.dbc writes a separate file and leaves the input unchanged. This is the safest choice for destructive or experimental edits.--no-backup disables the .bak copy for an in-place save. Use it only for disposable files or when another backup exists.--encoding with the encoding used by the source and destination, such as cp936 for a compatible legacy file.0 through 0x7FF; extended frame IDs: 0 through 0x1FFFFFFF with --extended.--fd: 1–64 bytes.SG_MUL_VAL_ metadata.0|0 is reserved for absent bounds; floating-point zero is representable. See the field reference before clearing or setting bounds.python3 scripts/dbc_crud.py --json excel export INPUT.dbc OUTPUT.xlsx
python3 scripts/dbc_crud.py --json excel import INPUT.xlsx OUTPUT.dbc
Metadata, Nodes, Messages, Signals, Choices) with dropdown validation for boolean and byte-order columns; an existing destination is replaced atomically without a backup..xlsx files and rejects formula cells — paste computed cells as values before importing. The whole workbook is validated before anything is written, so a failed import never touches OUTPUT.dbc; a successful import replaces it atomically and backup is always null.result.warnings.compare the two files and check result.equal.The full workbook schema, column list, and cell syntax are in the CLI reference; result keys are in the field reference.
The workflows use a temporary copy and write every mutation to a new --output file. They demonstrate CRUD, scale editing, standard and extended multiplexing, safe deletion, comparison, a semantic Excel round trip, final inspection, and validation.
.venv/bin/python -m unittest tests.test_docs -v
.venv/bin/python -m unittest discover -s tests -v
git diff --check
Released under the MIT License.