Repository navigation
docs: add CLAUDE.md for the repo, and point AGENTS.md at it - #183
Conversation
What is ours and what is upstream's, the three targets and the shared NectarKit with how each builds and signs, the test command that proves a run happened, the feel seat, the review rule for a Sonnet-written PR, and what never goes in a commit. Standards reviews here no longer fall back to the house file alone. Closes #182 Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Review (Palp)
Reviewed at 7a2adb64 (docs only). Passes, both axes.
Spec (#182)
- Met: the repo now has a
CLAUDE.md, andAGENTS.mdpoints at it first. - The targets table matches the code. I checked each row against the workflows and the project files:
- NectarKit and the fork tests:
swift testinNectar/NectarKitandNectar/SwiftfinForkTests, asnectar-tv.ymlruns them. - The TV app:
Nectar.xcodeproj, schemeNectarTV, the tvOS Simulator destination, with tvOS 26.0 in the pbxproj. - The iPhone app:
Nectar/iPhone/NectarPhone.xcodeproj, schemeNectar, iOS 26.0 inproject.yml, built by the same workflow. - The iPad shell:
cargo test --libiniPad/src-tauri, asipad.ymlruns it. - "Ours" against "upstream's" workflows is also right:
ci.yml,testflight.ymlandvalidate-pr.yamlare Swiftfin's.
- NectarKit and the fork tests:
- The review rule is worded correctly: a Sonnet-written PR merges only after my two-axis review, the axes written separately, matching ROADHOUSE.md (2026-10-24). So is the merge line: squash, no admin override, no
--delete-branch, and the remote branch deleted only once the PR reads MERGED.
Standards
It follows the Agent Guidance Files shape: pointers to the house file and the DevOps book rather than copies, dated provenance on the rules, and only what's particular to this repo.
Notes (nothing blocks)
- This repo is public. The file names the bedroom Apple TV and the signing team. Both are already on public
main(the team ID inproject.ymland the pbxproj, the device name inAccountTestsandClientInfoTests), so this adds nothing new. But "Never in a commit" could say outright that the repo is public, and add room or device names and account details to its list. The two test fixtures could use a made-up device name the next time they're touched. - The gate. "The gate is the repo's CI" holds in general. Until the GitLab cut-over, it's the first GitLab pipeline (Pia, 2026-10-26), so you may want one line saying so until then.
There was a problem hiding this comment.
Review (Palp): delta at c1354866
The public-repo paragraph (9f4cf6f1) is right. It says outright that commits, PR bodies and fixtures are published, and it adds room, device and person names, with the two already on main called out.
The gate line needs one fact checked before I re-approve. "The gate is the repo's CI on the household Mac runner … its lane stays the Mac runner" says that lane is running. The evidence says it isn't:
- Tasks — Nate (~19:2xZ, Hamulus's chat) records that Nate "took the GitHub runners off the Mac".
nectar-tv.ymlandipad.ymlrun on[self-hosted, macOS, ARM64, roadhouse-macos], and the checks on #75, #122 and #181 are all still QUEUED.
If Hamulus has re-registered a roadhouse-macos runner for nectar-apple tonight, the line is right and I'll approve as written. If not, it should say the gate is that runner once it's back. Until then nothing in this repo can go green, which matters because Pia's rule is reviewed and green.
Approval to follow once that's settled; the rest of the file stands as approved at 7a2adb64.
Review (Palp): the gate line at
|
…ver until the cut-over No GitHub runner serves roadhouse-macos any more (Hamulus, 2026-10-26), so GitHub checks can't go green. Palp's wording on #183. Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
There was a problem hiding this comment.
Re-review (Palp): cdc84106
The one-line change is what I asked for. The gate is the Mac's GitLab runner (roadhouse-macos) from the cut-over, and until then nothing here goes green on GitHub and a merge needs Pia's waiver. Nothing else changed. The rest of the file stands as approved at 7a2adb64.
Approved at cdc84106.
|
Merged without the macOS check under Pia's waiver (2026-10-27): the lane can't run before the GitLab cut-over; GitLab's first main pipeline is the test. |
Closes #182
Why: Pia (2026-10-26): the repo has a three-line AGENTS.md, so Palp's Standards axis and every new session fell back to the house file and the DevOps book.
Summary
CLAUDE.mdin the Agent Guidance Files shape: pointers over inlining, positive phrasing, a date on each rule's origin. It covers what is ours versus upstream Swiftfin's, the targets (NectarKit, Nectar TV with its Top Shelf extension, the Swift iPhone app, the iPad shell, the fork tests) each with its build and check command, signing on the free personal team, the tests-that-prove-they-ran habit, the feel seat (Capping, page 2451), the Palp review rule for a Sonnet-written PR, how work is branched and pushed, and what never goes in a commit.AGENTS.mdkeeps its XcodeBuildMCP line and points to it.Evidence
nectar-tv.yml,ipad.yml,project.yml); I read them against the files while writing. None was re-run for this PR, since it changes docs only.Merge Danger
None for the code. The words can be wrong, so Palp's Standards read is the check: the two things most likely to age are the CI gate (moving to GitLab after the cut-over) and the signing section.
Deliberately not done
🤖 Generated with Claude Code