[Bre77]

This PR revises AGENTS.md against the last month of agent sessions in this repository (14 sessions since 2026-10-03) and the current state of main. It changes no code.

Correction, buf breaking entry under "Proto content rules": the old entry said a renumber is never possible and that the only sanctioned fix is to reserve the old name and add a new field. Seven sessions shipped or prepared approved breaking corrections another way, with a temporary breaking.ignore_only block in buf.yaml (#68 added one for common_api.proto, #70 deleted it after 4.0.0, #75 replaced the 4.0.1 telemetry block with one for three command files, and the field 272-278 resync, the 5.0.0 follow-up and the 6.0.0 full pass each built or merged one). One more session stopped for a decision because nothing told it this path existed, and a review on #75 objected that a buf.yaml change loosened checks beyond the approved renames. The entry now describes the mechanism: a maintainer decision first, then a block that names only the rules that fired and the files they fired on, merged with any existing block and deleted in the first PR after the release. The reserve-and-rename option and the NavigationSuperchargerRequest example stay, because car_server.proto points to them.

Correction, upstream PR pin under "Upstream sync": the old entry limited pinning to a PR on a branch the upstream org owns, not a fork. Two sessions found the upstream PRs they had to pin come from a fork (fleet-telemetry #572 and #582), pinned them anyway, and the pinned gate passed, because the upstream repo's raw URL serves the fork's head commit. main pins #582's fork head today. The entry now says forks work. It also adds one clause about the weekly upstream-drift.yml PR, which moves a PR-head pin back to upstream main and drops that PR's symbols. One session was told to fix that PR (#65) on its bot branch before it found that every run force-pushes upstream-reconcile, and the same bot PR is open again now (#69).

Addition, .venv rule under "Generated code": four sessions ran scripts/build_catalog.py, scripts/upstream_coverage.py or a Python check with system python3, or before .venv existed, and hit ModuleNotFoundError: No module named 'google'. One of them reported the coverage gate as impossible to run, and one built a separate virtualenv. One session also hit buf: command not found. The new line says to use .venv/bin/python, that pnpm install && pnpm generate creates it, and to run buf through pnpm exec.

Correction, gencode stamp entry: it said a pin bump rewrites "all 38 _pb2.py files"; there are 44 now, so the text says "every _pb2.py file".

Removal, the publish.yml bullet under "Release": it restated README "Releases", which already explains that publish.yml reuses ci.yml and that both publish jobs wait on it. No session touched the release path. The pointer to README for publish flow and interrupted-publish recovery stays in one shorter line.

Not changed here, because it is code: the temporary ignore_only block in buf.yaml and the matching --against-config buf.yaml flag in ci.yml both say to delete them in the first PR after 6.0.0 is published, and 6.2.0 is now out.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

Run configuration

• Source mode: Auto
• Repository: Teslemetry/tesla-protocol
• Branch: fm/agentsmd-hist-tesla-protocol
• File: .coderabbit.yaml
• Commit: 8afc6da
• Layers: Repository YAML
• Review profile: CHILL
• Plan: Team
• Run ID: 87d55c00-aee6-4423-84dc-e0a18b89fab6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

CodeRabbit Verify

• Run CodeRabbit Verify to test the changes in this pull request.