autosar-agent-skills
Agent skills for AUTOSAR Classic Platform engineering. Zero third-party
dependencies — every script is standard-library Python, so it runs inside an
air-gapped build server, a CI container, or an agent sandbox without a network
round-trip or a pip install.
Every skill here is self-contained: drop the directory into an agent's skill path and the agent loads it, or run the script directly as a CLI tool.
They all target one class of engineering problem: bugs that compile cleanly and only surface during ECU integration testing — a signal decoded three bits off, a frame length that changed without anyone updating the config, a receiver that silently drops every frame because a CRC seed moved.
Skills
| Skill | What it does | Deps |
|---|---|---|
| dbc-matrix-audit | DBC communication matrix audit: bit-field overlap, out-of-range signals, left/right asymmetry, naming drift, GBK encoding | none |
| arxml-dbc-consistency | ARXML comm-stack × DBC cross-validation: binding, DLC, bit geometry, byte order, init values, PduR routing, TX/RX direction | none |
Also listed on the Agensi marketplace: DBC Matrix Audit (free).
The English packages uploaded there are generated from this repo by
tools/pack_marketplace.py; the same input produces byte-identical output.
Note:
LISTING.mdrecords the sha256 of the zip as submitted. Editing a skill's text afterwards (e.g.8b27f44fixed an example README) changes the rebuild hash — that is content changing, not broken reproducibility. To verify, build on the corresponding commit first, then compare.
dbc-matrix-audit
Static auditor for DBC communication matrices: 11 structural checks, two-revision diff, live frame decoding, and CSV export — standard library only.
python skills/dbc-matrix-audit/scripts/dbc_analyze.py info matrix.dbc
python skills/dbc-matrix-audit/scripts/dbc_analyze.py messages matrix.dbc
python skills/dbc-matrix-audit/scripts/dbc_analyze.py lint matrix.dbc
python skills/dbc-matrix-audit/scripts/dbc_analyze.py decode matrix.dbc 0x1D2 1234052a00000000
python skills/dbc-matrix-audit/scripts/dbc_analyze.py export matrix.dbc --csv inventory.csv
python skills/dbc-matrix-audit/scripts/dbc_analyze.py diff old.dbc new.dbc
Checks
| Code | Severity | Meaning |
|---|---|---|
ENC-001 |
error | File is not UTF-8 (typically GBK comments) |
BIT-001 |
error | Two signals in one frame occupy the same bit |
RNG-001 |
error | Signal extends past the frame length |
DLC-001 |
error | Illegal DLC |
ENU-001 |
warn | ENUM attribute assigned a numeric index |
SYM-001 |
warn | Left/right signal sets (L/R, A/B, FL/FR…) are asymmetric |
TYP-001 |
warn | Same signal spelled differently once the side token is removed |
DUP-001 |
warn | Signal name reused across frames |
STA-001 |
info | Many signals have no GenSigStartValue |
NOD-001 |
info | Transmitting node not declared in BU_ |
VAL-001 |
info | Short signal (≤4 bit) has no value table |
Revision diff
| Code | Severity | Meaning |
|---|---|---|
DIF-GEO |
error | Signal bit definition changed |
DIF-DLC |
error | Frame length changed |
DIF-REN |
error | Frame / signal / node renamed |
DIF-ID |
error | Frame with the same name moved to a different CAN ID |
DIF-CRC |
error | DataID inside the checksum comment changed |
DIF-DEL |
warn | Frame or signal deleted |
DIF-SCL |
warn | Factor / offset changed |
DIF-RNG |
warn | Min / max / unit changed |
DIF-VAL |
warn | Value table changed |
DIF-CYC |
warn | Cycle time changed |
DIF-TX |
warn | Transmitting node changed |
DIF-TYP |
warn | Send type changed |
DIF-ADD |
info | Frame or signal added |
DIF-CMT |
info | Comment-only change |
Two miss patterns handled deliberately:
- Same name, new ID. Comparing by ID alone degrades into "one deleted + one added", burying the actual cause. The tool pairs by name first and suppresses the misleading add/delete pair.
- CRC DataID change. It hides inside a
CM_comment (CRC16 checksum (DataID: 0x61D2)), so a plain comment diff reportsinfo. The tool regex-extracts it and escalates toerror— every receiver will reject the whole frame.
Renames are matched by bit definition (the name is precisely what changed),
so Bus_Off_Count → Bus_Off_Cnt is one DIF-REN, not a delete plus an add.
# Real case: NM frame moved to a new ID + CRC DataID changed — either one breaks the link
$ python skills/dbc-matrix-audit/scripts/dbc_analyze.py diff old.dbc new.dbc
[E] DIF-CRC Node_Info_Checksum: checksum data id changed (0x1D2)
0x61D2 -> 0x41D2. Every receiver that validates this CRC will reject
the frame until it is rebuilt with the new id.
[E] DIF-ID NM_Autosar_ECU_R: CAN id changed (0x457 -> 0x46B)
Same name, different address: routing tables and filters still point
at the old id.
Rule details and false-positive scenarios: references/checks.md.
Why not cantools
cantools is a good library, but parsing throws away exactly what an audit needs:
| Data | cantools | This tool |
|---|---|---|
Cycle time GenMsgCycleTime |
✗ | ✓ |
Value tables VAL_ |
partial | ✓ |
Comments CM_ |
✗ | ✓ |
Frame/signal attributes BA_ |
✗ | ✓ |
cantools also silently tolerates GBK byte sequences — parsing appears to
succeed while every comment is lost. This tool reports ENC-001 instead.
Example
$ python skills/dbc-matrix-audit/scripts/dbc_analyze.py lint skills/dbc-matrix-audit/examples/demo.dbc
[I] STA-001 165/167 signals have no GenSigStartValue
...
summary: 0 error, 0 warn, 1 info
$ python skills/dbc-matrix-audit/scripts/dbc_analyze.py lint skills/dbc-matrix-audit/examples/demo_broken.dbc
[E] BIT-001 ACM_B_Supply_Voltage and ACM_B_Temperature occupy overlapping bits (0x201 ACM_B_Status)
[E] DLC-001 illegal DLC 13 (0x500 Log_Snapshot)
[E] ENC-001 file is not valid UTF-8 (detected: gbk)
[E] RNG-001 Log_Word_4 extends past the frame length (0x500 Log_Snapshot)
[W] DUP-001 signal name Node_Status_Counter appears in 2 frames
[W] ENU-001 1 ENUM attribute assignments use numeric indices
[W] SYM-001 ACM_A_Status / ACM_B_Status are not symmetric (0x200 vs 0x201)
[W] TYP-001 same signal spelled differently (beyond the side token)
ACM_X_Projection_State vs ACM_X_Projection_statE
summary: 4 error, 4 warn, 2 info
examples/demo.dbc is a clean synthetic matrix; demo_broken.dbc plants 10
classes of defect for demonstration. Both are generated by tools/gen_demo_dbc.py,
which computes bit allocation programmatically — they cannot structurally overlap.
arxml-dbc-consistency
Cross-validates AUTOSAR Classic ECU configuration (Com / PduR / CanIf / CanDrv ECUC containers) against a DBC communication matrix. 26 check codes, standard library only.
python skills/arxml-dbc-consistency/scripts/arxml_dbc_check.py probe ecuconfig.xml
python skills/arxml-dbc-consistency/scripts/arxml_dbc_check.py check ecuconfig.xml matrix.dbc
python skills/arxml-dbc-consistency/scripts/arxml_dbc_check.py matrix ecuconfig.xml matrix.dbc --csv inventory.csv
python skills/arxml-dbc-consistency/scripts/arxml_dbc_check.py selftest
probe inspects structure first (vendor container names, parameter names, and
where signals are mounted vary enormously); check then compares. check exits
1 on error, so it drops straight into CI.
Why name matching cannot work
In a real configuration Com and CanIf are not directly connected; the only sound binding key is the CAN ID:
DBC frame ──CanId──> CanIf PDU ──CanIfXxxPduRef──> EcuC Pdu(X)
──PduR routing──> EcuC Pdu(COM_X)
──ComPduIdRef──> Com IPdu
IPDU_COM_TP_Param has nothing name-wise in common with the DBC frame
Diag_CalData, yet they are the same link
(COM_TP_Param → CanTp_TP_Param → 0x9B). Joining by name is guaranteed wrong.
There is exactly one correct bit-geometry formula
AUTOSAR's ComBitPosition always stores the signal LSB (flat numbering,
bit0 = LSB of byte0), regardless of byte order. DBC numbers Motorola signals
in sawtooth form and stores the MSB, so first walk back len-1 steps:
def motorola_msb_from_lsb(lsb, length):
pos = lsb
for _ in range(length - 1):
pos = pos - 15 if pos % 8 == 7 else pos + 1
return pos
Measured on a production project: 215/215 signals matched. An earlier version assumed "MSB and LSB schools coexist", tried both, and accepted either — that design is wrong: a genuine error gets whitewashed by a coincidental match, and the measured hit rate was only 153/215.
A whole-network matrix is not one ECU's view
A DBC may describe only this ECU (a dozen frames), or the entire network
(hundreds of frames, with many other nodes under BU_:). In the latter, most
frames belong to other parties and this ECU should have no configuration for
them at all. The script decides frame by frame — "sender is this ECU, or some
signal's receiver includes this ECU" — and only raises error on relevant frames.
On an 11-node, 91-frame matrix, the old "every frame must be configured" assumption produced 99 false errors and buried all 6 real defects.
Bus coverage
An ARXML often drives several CAN buses at once. The script walks back through
CanIfBufferCfg → CanIfHthCfg → CanIfHthCanCtrlIdRef to identify the bus and
computes each bus's hit rate against the DBC: ≥20% is compared normally, <20% is
skipped wholesale with an info. The check header prints it, and reports must
quote it verbatim:
bus CANIF_CTR_BCAN 16/21 compared | CANIF_CTR_PCAN 0/32 skipped
Reference docs
- Chinese:
skills/arxml-dbc-consistency/references/checks.md - English:
skills/arxml-dbc-consistency/references/checks.en.md - English SKILL.md (for marketplace packages):
packaging/agensi/arxml-dbc-consistency/SKILL.md
Documentation site
Rendered docs with full schema markup: https://autosar-skills.github.io/autosar-agent-skills/
Install
git clone https://github.com/autosar-skills/autosar-agent-skills.git
cd autosar-agent-skills
python -m unittest discover -s tests # 65 tests, standard library only
No pip install required — the scripts run from the clone as-is. If you want
the short commands on your PATH:
pip install dbc-matrix-audit
dbc-analyze lint matrix.dbc # DBC auditor
arxml-dbc-check check cfg.arxml matrix.dbc # ARXML x DBC cross-check
Note:
dbc-analyzeoutput and help are English.arxml-dbc-checkcurrently emits Chinese messages — the audit vocabulary (bit geometry, PduR routing, FD flags) is identical, but the prose is not translated yet.
Using as an agent skill
skills/<name>/SKILL.md follows the general AgentSkill format. Put the whole
skills/<name>/ directory on the agent's skill path; scripts are invoked by the
relative path scripts/dbc_analyze.py.
SKILL.md is a runbook for the agent (mode detection → execution order → output interpretation), not human documentation. This README is the human-facing half. The split:
| Location | Reader | Contents |
|---|---|---|
README.md / README.en.md |
humans | Why it is designed this way, what rules mean, sample output |
SKILL.md |
agent | Step 0–4 workflow, commands, reporting format |
references/checks.md |
loaded on demand | Per-rule decision detail and false-positive scenarios |
OpenClaw users:
openclaw skills install skills/dbc-matrix-audit # first install
openclaw skills install --force skills/dbc-matrix-audit # re-sync after repo changes
openclaw skills info dbc-matrix-audit # confirm Ready
Installation copies the directory into ~/.openclaw/workspace/skills/, so a
repo update requires re-running with --force.
Repository layout
skills/dbc-matrix-audit/
├── SKILL.md agent runbook (Chinese: mode / flow / reporting)
├── scripts/dbc_analyze.py implementation (single file, standard library)
├── examples/ synthetic fixtures (clean / seeded), shipped with the skill
└── references/ rule detail, encoding notes, DBC syntax
skills/arxml-dbc-consistency/
├── SKILL.md bilingual runbook (mode / flow / reporting)
├── README.md English description (repo and marketplace)
├── scripts/arxml_dbc_check.py implementation (single file, standard library)
├── scripts/demo_fixtures.py program-generated fixtures (topology must match production)
├── scripts/test_contract.py bidirectional assertion: documented codes ↔ emitted codes
└── references/ checks.md (zh) / checks.en.md (en)
tools/gen_demo_dbc.py fixture generator
tools/pack_marketplace.py marketplace packager (reproducible, structure-checked)
tests/ 65 unit tests (including diff-specific)
packaging/agensi/ listing material: English SKILL.md / README overlays, copy archive, logo
CHANGELOG.md version history
dist/ build output (gitignored, regenerate with the scripts above)
Marketplace packaging
python3 tools/pack_marketplace.py # → dist/agensi/dbc-matrix-audit.zip
python3 tools/pack_marketplace.py --skill arxml-dbc-consistency # → dist/agensi/arxml-dbc-consistency.zip
python3 tools/pack_marketplace.py --check # validate without writing
SKILL.md / README.md under packaging/<market>/<skill>/ override the Chinese
originals; every other file comes from skills/<name>/. The flat layout
packaging/<market>/ is still supported for older skills — only its top-level
files are read, so a sibling skill's overlay cannot leak in. The packager strips
__pycache__, binaries, dotfiles and symlinks, and validates "single top-level
folder + SKILL.md at its root + README present", failing otherwise. Entries are
sorted with fixed timestamps, so repeated builds are byte-identical.
License
MIT. All sample DBC files are fictional and contain no real vehicle model, supplier, or personnel information.
Background
Design rationale and the debugging story behind these tools: How I designed an engineering-grade skill for AI agents · 105 errors, 99 of them fake
Metadata
Release files for dbc-matrix-audit 1.3.2
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| dbc_matrix_audit-1.3.2.tar.gz | 61.4 kB | Without provenance | Details for dbc_matrix_audit-1.3.2.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| dbc_matrix_audit-1.3.2-py3-none-any.whl | Python 3 | none | any | Details for dbc_matrix_audit-1.3.2-py3-none-any.whl |
Total release size: 116.2 kB
Release files / dbc_matrix_audit-1.3.2.tar.gz
| Download URL | dbc_matrix_audit-1.3.2.tar.gz |
|---|---|
| Size | 61.4 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
63cdf601d63670044df57ddd8f7b30df96855a3cc3e9dc2f099484784945ec2c
|
|
BLAKE2b-256 checksum How to use checksums |
9553d93dcb1671a3aa5e5654b30ba3f4b32209df897f1bb4ea097a1f2b340a2e
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.3
|
Release files / dbc_matrix_audit-1.3.2-py3-none-any.whl
| Download URL | dbc_matrix_audit-1.3.2-py3-none-any.whl |
|---|---|
| Size | 54.8 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
ffa0bdbb6a81c845bf3f0b1fd47b3ec698839dbbc74829da816f7ddead3e9c06
|
|
BLAKE2b-256 checksum How to use checksums |
4ca04a8a99f265becb4b25c4e494855586aaaa5642d90ed9f8e476af58067a0b
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/7.0.0 CPython/3.12.3
|
Security for dbc-matrix-audit 1.3.2
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.