Move Project to GitHub with CI

The repo is now on GitHub with an Entire mirror, and CI is set up. The trail move is blocked, and the CI fix is still running.

What's in place

• GitHub: https://github.com/petercr/orca-entire-extension (public). I pushed all four branches (main, auto-refresh, plugin-scaffold, vps-workspaces) and the 16 Entire checkpoint refs.
• Remotes: origin now points at GitHub. The old Entire repo is still there as a remote named entire-old, untouched. Your four branches track origin.
• Entire mirror: /gh/petercr/orca-entire-extension is ready on aws-us-east-2. entire now finds gh/petercr/orca-entire-extension from origin, so new trails go to project gh/petercr.
• CI: https://github.com/petercr/orca-entire-extension/pull/1 adds .github/workflows/test.yml. It runs node --test on Windows and Linux with Node 24, after scripts/create-panel.mjs.
  • The first run passed on Windows and failed on Linux. isInstalledCopy in src/worker/panel.mjs used the platform's path module, which on Linux doesn't split Windows \ paths. I switched it to path.win32, which splits on both.
  • All 63 tests pass locally. The new CI run had just started when I last checked, so the Linux result isn't in yet.

Blocked: moving trail #3 Entire refused to create the new trail: "repoIds must contain repositories owned by the project with Trails enabled (422)". The CLI has no setting for this, so Trails probably has to be turned on for the new repo on entire.com. It might also start working once Entire finishes processing the new mirror. Once Trails works, I'll create the new trail and close old trail #3 (https://entire.com/et/entire-repos/-/trails/3) with a link to it. I've left #3 open until then. Its vps-workspaces change already shows as merged.

Not touched

• Your uncommitted SSH-targets work is still on main, uncommitted.
• I didn't edit CLAUDE.md or README.md because you have uncommitted edits in CLAUDE.md. The README's "Try it" steps still hold for a GitHub clone.
• If you want to remove the old Entire-native repo later, run entire repo delete /et/entire-repos/orca-entire-extension yourself. Its three trails live in et/entire-repos, so deleting it may affect them.

The trail move is done:

• New trail: https://entire.com/gh/petercr/-/trails/378 holds the VPS work, including the SSH reads you haven't committed yet. It has no branch linked yet.
• Old trail: https://entire.com/et/entire-repos/-/trails/3 is closed, and its description now starts with a link to #378.

CI on https://github.com/petercr/orca-entire-extension/pull/1 now passes on Windows and Linux.

Setting up the code review trail

Entire's code review is entire review. Several reviewer agents review the branch in parallel, then a judge merges their reports. With the output set to trail, the verdict is posted on the branch's trail as findings, so it shows in Orca's Entire panel. This repo has no review profile yet. Setting one up is your command to run, so here is one you can paste into PowerShell:

• Reviewers: Claude Opus and Codex review independently, and Opus judges. These are the review agents your CLI offers; Pi is the other one.
• Shared profile: without --local, the profile goes into .entire/settings.json, which is committed, so commit it and every clone gets it. Add --local to keep it to yourself.
• Wizard: to choose these settings interactively, run entire review --configure with no other flags.

Running a review

The review needs a branch that is linked to a trail. For the SSH work:

You can also review from anywhere with entire review --target https://entire.com/gh/petercr/-/trails/378. It checks the branch out in a separate worktree first.

If you'd like me to run the configure command, commit the profile, or put the SSH work on that branch, say so.

I pointed you at the wrong feature before: entire review is the local multi-agent review. The cloud setting you mean is the project's review agents. Entire runs them on each push and posts their findings on the trail. They're managed with entire project agent, and the web setting is the same thing.

Your gh/petercr project already has two of them, but neither covers this repo yet. This repo's Entire ID (01M6CDEHGQ…) isn't in either agent's repo list:

Agent	Preset	Runs on	Repos now
Code review	code-review@1	Codex, gpt-6.1-sol	6 others
Security review	security-review@3	Claude Opus 5.5	2 others
Slop	slop	Claude Opus 5.5	not added to the project

To add this repo, run this from the repo folder. Project setup is yours to run, so I haven't run it:

Each preset runs only once per project, so this adds orca-entire-extension to the agents you already have. It doesn't create new agents, and their models, triggers and environment stay the same. Add slop to the list if you also want that rating.

I'd include security review here. The worker runs entire, wsl.exe and ssh with text that comes from Entire and Orca, which is the kind of injection path that agent traces.

When they run

• They run on a push to a branch whose Change is ready for review.
• With a GitHub PR, they don't run while it's a draft. They start once it's marked ready.
• So for the SSH work: push a branch, link it to trail #378 with entire trail link 378 --branch <branch>, and open a PR that isn't a draft.

Which rules they follow

• The code-review agent reads REVIEW.md, AGENTS.md and CLAUDE.md as they are at the diff base. The nearest file to a changed file wins.
• Your CLAUDE.md rules will be applied, such as treating Entire and Orca text as untrusted data and the 30-second budget.
• If you want review-only rules, add them in a REVIEW.md at the repo root.

Afterwards, entire project agent list should show a higher repo count for each agent you added. I can write a REVIEW.md if you'd like.

Confirmed: Code review now covers 7 repos and Security review 3, and both include orca-entire-extension and are enabled. You didn't add Slop.

They run only on a branch that has a Change on a trail, which needs entire trail link, so neither is reviewing anything yet:

• CI PR: https://github.com/petercr/orca-entire-extension/pull/1 has no trail, so they'll skip it. CI passed on Windows and Linux, so you can merge it as is.
• SSH work: your uncommitted SSH-targets work is the first real candidate. Put it on a branch, push it, run entire trail link 378 --branch <branch>, and open a PR that isn't a draft. Both agents will then post their findings on https://entire.com/gh/petercr/-/trails/378.

I can do that branch, commit, link and PR for you if you'd like.

The SSH work is on its own branch, linked to the trail, with a ready-for-review PR open: https://github.com/petercr/orca-entire-extension/pull/2

PR #1 isn't merged. GitHub still shows https://github.com/petercr/orca-entire-extension/pull/1 as open, and main doesn't have it. So main still lacks the CI workflow and the Linux panel fix, and CI won't run on PR #2 until #1 is merged.

What I did

• Created branch vps-ssh from main and committed all your uncommitted SSH work in one commit (18d6fdf). That covers ssh-targets.mjs, the hosts/snapshot/process changes, the panel text, the tests, and the CLAUDE.md and PLAN.md updates. All 63 tests pass.
• Replaced your VPS's real IP in the two test files with a documentation address, 203.0.113.10, before pushing to the public repo. The real IP and user (ubuntu@15.204.88.195) are still in docs/PLAN.md line 14, which was already public on GitHub when I pushed main earlier. Say if you want that line made generic. It would still be in the git history.
• Pushed the branch and linked it to trail #378 as change orca-entire-extension/1: https://entire.com/gh/petercr/-/trails/378/changes/orca-entire-extension/1

The PR Linking the trail made Entire open a draft PR for the branch, so that became PR #2 instead of one I created. I gave it a real title and description, kept Entire's hidden marker (Entire uses it to close the PR when the trail closes), and marked it ready for review.

The cloud reviews

• The trail shows no review results yet: approvals are failing (no one has approved), checks haven't been evaluated, and findings pass with nothing blocking.
• The review agents run on pushes, and I pushed while the PR was still a draft. If Code review and Security review don't show up on the trail, the next push to vps-ssh should start them.

I fixed the Code review finding and pushed it to the PR branch (commit 5733329). The push should start a new cloud review on https://entire.com/gh/petercr/-/trails/378.

The bug: serverSshTarget saved each server's SSH target in a module-level map for the life of the worker. The worker now runs until Orca quits. So if you edited the target or re-paired the server, refreshes kept connecting to the old address until the plugin was turned off and on.

The fix: in src/worker/ssh-targets.mjs I removed the cache, so the target is looked up again on every refresh. A server refresh already makes several orca CLI calls, and this adds one more plus a read-only SQLite query. It runs only when a refresh happens, not on the 5-second focus poll. I added a code comment explaining why it isn't cached.

Test: the two lookups can now be swapped for fakes in tests. A new test changes the saved address, switches the server to a different target, and then removes the target. It checks that each lookup picks up the change, which the old cache would have failed. All 64 tests pass.

PR #1 still isn't merged, so GitHub CI won't run on https://github.com/petercr/orca-entire-extension/pull/2 yet.

PR #1 is merged, and I've brought main into the SSH branch so CI now runs on https://github.com/petercr/orca-entire-extension/pull/2.

• Your local main: fast-forwarded to 258eb06 (the merge of #1).
• SSH branch: I merged origin/main into vps-ssh (36e5434) and pushed. It now has the CI workflow and the Linux panel fix. There were no conflicts, and all 64 tests pass locally.
• Checks on PR #2:
  • Socket Security passed.
  • The Windows and Linux test jobs and Entire's gates for trail #378 were still running when I checked.
  • The earlier CI run took about 20 seconds.
  • This push should also start Entire's new Code review and Security review runs on https://entire.com/gh/petercr/-/trails/378.

I fixed the second Code review finding and pushed it to vps-ssh (commit 1602068). The push should start new cloud reviews on https://entire.com/gh/petercr/-/trails/378 and CI on https://github.com/petercr/orca-entire-extension/pull/2.

The bug: when Orca saved no port for an SSH target, sshTarget filled in 22, and the ssh command always passed -p. A -p on the command line overrides the Port set in ~/.ssh/config, so an alias using another port couldn't connect.

The fix:

• src/worker/ssh-targets.mjs now leaves the port unset when Orca saved none. It still rejects a saved port that isn't a whole number from 1 to 65535.
• src/worker/hosts.mjs passes -p only when there is a saved port. Without one, ssh uses the alias's Port, or 22.
• I added tests for an alias without a port and for the ssh arguments without -p. All 65 tests pass.

One gap: if Orca saves a port for every target, including one taken from ~/.ssh/config, then -p still overrides the alias. I don't know how Orca stores those. Your VPS target is a manual entry with port 22, so it isn't affected.