Set up Chess-specific Claude Code and Herdr engineering agents

[nicksan222]

Summary

Add an optional Claude Code / Herdr engineering team alongside the existing Pi setup, adapted from nicksan222/study but scoped to this physical chessboard.

• Eight default roles: lead, portable Rust developer, firmware engineer, hardware engineer, mechanical engineer, QA, reviewer and pushback. Manufacturing, test engineering, DevOps, PM, upgrade review and PR delivery are on demand; maximum ten active agents. No student or generic scenario role.
• Project-specific briefs define PCB/power/SPICE, CAD tolerances/optical stack, Pi Linux adapters, manufacturing evidence and Yocto/toolchain responsibilities, with explicit shared-contract handoffs and one writer per file.
• Pin Claude Code 2.1.287, checksum-verify Herdr 0.9.3, install Codex 0.159.3 and optional Reviewr v0.39.0. Persist runtime/login state in separate Docker volumes without mounting host credentials or embedding secrets in the image.
• Add lifecycle, setup, doctor, token-reporting and validation recipes. Permission bypass is opt-in; setup does not sign in, trust the checkout, start models or publish work.
• Bind workspaces to the checkout, isolate explicit sessions from inherited socket/machine context, retain blocked startup panes, read native API failures from stderr and enforce the cap across incremental starts. Optional Reviewr outages do not block container setup.
• Add offline CI coverage and include agent lint/format/tests in the normal repository gates. Preserve Pi's configuration and storage.

No visual demo: this changes contributor tooling and agent instructions, not application or hardware appearance. No hardware design changes, fabrication exports or physical validation claims are included.

Validation

• devcontainer build --workspace-folder . --image-name chess-agent-review:local: full image build passed, including pinned native Claude and checksum verification of Herdr.
• Executed .devcontainer/post-create.sh twice in that image with isolated empty agent volumes and no forwarded provider credentials: provider hooks, Reviewr/Chess plugin installation, CLI versions, frozen Pi install/checks and roster passed.
• HERDR_TEST_BIN=<Herdr 0.9.3> just agents-check: 42 tests passed, including three native tests using temporary repositories/isolated servers; verified checkout/session ownership, plugin socket/workspace validation and stderr API errors. No provider was launched by these tests.
• Repeated the native suite three times and ran the complete offline suite with both container markers simulated absent: host CI portability and the real host-rejection guard passed.
• Native Claude parsed the launcher flags successfully; doctor rejected the unauthenticated setup with the documented login guidance. No authenticated model request was made; login/trust remain explicit user actions.
• bun run --cwd .pi check: TypeScript check and 32 tests passed, both locally and during container setup.
• Ruff, ShellCheck, Just formatting and Git whitespace checks passed.
• Commit passed the unmodified Git hook invoking just precommit: Rust/package checks, firmware AArch64 linkage, shared Python, CAD fast checks and PCB review/SPICE/ERC/DRC. Incidental regenerated hardware artifacts are not included in this tooling PR.

Checklist

• The change is focused and includes relevant tests.
• Documentation is updated where behavior or workflows changed.
• I ran the relevant package recipe and just precommit through the normal commit hook.
• Hardware claims distinguish automated validation from physical evidence.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: f5f7c1c7-8847-4adb-a853-d340d5f84797)

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-04T12:34:52.129953Z	3905069	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[cursor]

Approved. Cursor Bugbot was skipped, so that signal was not used; no applicable approval policy required human review, and there were no unresolved automated-review findings that need attention.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 39050692ce

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Mock the devcontainer guard in the offline setup test

On the ubuntu-latest host used by both newly added agent-team workflow jobs, this call immediately raises SystemExit("run setup inside the devcontainer") because neither container marker exists. Running the workflow's exact unittest command reproduces the failure in this environment, and agent-team is included in the PR's required-checks, so every PR is blocked before the mocked Reviewr behavior is tested; mock the container check or test the setup logic below that guard.

AGENTS.md reference: AGENTS.md:L22-L27

Useful? React with 👍 / 👎.

[nicksan222]

Fixed in e9ed021. The optional-Reviewr unit test now mocks require_container without weakening the production guard. A separate regression verifies that host setup exits before filesystem/tool mutations. The complete offline suite also passes with both container markers simulated absent; the native suite passed three consecutive runs.

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: 0cc4f8bc-5ed2-41e1-a742-09d83c07b418)

[cursor]

Approved. Cursor Bugbot was skipped, so that signal was not used; no applicable approval policy required human review, and there were no unresolved configured automated-review findings that need attention.

  

_{Sent by Cursor Approval Agent: Pull Request Router and Approver}