fix(platform): discover Nix agent CLIs - #6125
Conversation
|
Thanks @Pape45 — the shared I can see this is opened as a draft, so I'll hold off on a full review until you mark it ready. |
mrcfps
left a comment
There was a problem hiding this comment.
I verified that the shared toolchain resolver now includes both the standard per-user and system Nix profile bin directories, so the packaged PATH builder and daemon executable discovery receive the fix together without changing the existing precedence. The added regression assertions cover inclusion and system-bin opt-out behavior. I also ran all 80 platform tests, the platform typecheck, the repository guard, and the full workspace typecheck successfully. Thank you for the focused fix and clear verification—nice work! 🙌
🔁 Powered by Looper · runner=reviewer · agent=codex · An autonomous AI dev team for your GitHub repos.|
🧪 Queued for QA validation — this PR has changes that need a manual QA pass before it's merged. Nothing needed from you; we'll update here once it's validated. Thanks for the contribution! 🙏 |
|
Hi — I’m still actively maintaining this PR. It was marked ready on July 27, approved by @mrcfps, and then explicitly queued for manual QA with “Nothing needed from you.” I therefore waited for QA rather than posting activity-only comments. It was converted back to draft on July 30 for inactivity, and #6334 now proposes the same Nix paths. Could a maintainer please clarify whether #6125 should be restored to ready or considered superseded by #6334? I’m happy to follow whichever route you prefer. |
|
Heads-up: PR #6334 is also open against the same fix path — both PRs change Sharing this so the maintainer team can choose one canonical PR and neither thread ends up doing duplicate work. |
|
Thanks for the clear summary, @Pape45 — given the QA note on July 27, waiting here instead of posting activity-only comments was reasonable. PR #6334 is now open against the same fix path, so I’ve surfaced your question to the maintainer team as a restore-vs-supersede decision instead of asking you to churn on this branch in the meantime. Once they pick the canonical PR, we’ll follow that route here. |
Fixes #6121
Why
I hit this while launching Open Design 0.16.1 from Finder on macOS with Codex installed through Nix. The CLI resolves normally in an interactive terminal, but the packaged app inherits a stripped launchd PATH and therefore reports the agent as unavailable.
The shared toolchain resolver already compensates for Homebrew and common language/version-manager locations. Adding the standard user and system Nix profile bins closes the same GUI-PATH gap for both daemon executable resolution and the packaged sidecar PATH builder.
What users will see
Agent CLIs installed in
~/.nix-profile/binor/run/current-system/sw/binare discovered automatically by GUI-launched Open Design sessions. Existing search order and non-Nix installs are unchanged.Surface area
apps/weborapps/desktop(including Electron menu bar)odsubcommand or flag, newtools-dev/tools-packflag, or newOD_*env var/api/*endpoint, new SSE event, or changed shape inpackages/contractsskills/,design-systems/,design-templates/, orcraft/, or change to the skills protocolTRANSLATIONS.mdfor the locale workflow)package.json(dependenciesordevDependencies); workspace-packagepackage.jsonfiles are out of scope. Include a paragraph on what we get vs. what bytes we ship (seeCONTRIBUTING.md→ Code style)Screenshots
Not applicable; this changes executable discovery without adding or changing UI.
Bug fix verification
packages/platform/tests/index.test.tsmainand pass on this branch: yes.Validation
pnpm --filter @open-design/platform test -- index.test.ts— 80 passedpnpm --filter @open-design/platform typecheckpnpm guardpnpm typecheckAdjacent issues
#6122 is intentionally separate because exposing an unavailable Codex binary override is an independent Settings UI concern.