[kixelated]

Completes and deletes quest/m0/kix/github-fake.md, a follow-up of kixelated/quest#190.

Problem

cloudflare/test/publish.test.ts kept its own stateful GitHub fake, while the seeded local server's fake (cloudflare/dev/github.ts) modeled only the inbox flows. The local server could not run a job's publication, so the epic's end-to-end test could not start from the seeded fake.

Changes

• dev/github.ts now models what publication needs:
  • commits and trees in a fork network's shared object store, built only on known objects;
  • fast-forward (and forced) branch updates on both the upstream and its service fork;
  • pull request listing by head, creation (numbered after existing issues and pulls), title/body updates, and markPullRequestReadyForReview;
  • the issue and pull request reads behind a job's context (body, comments, reviews, threads, check runs).
• Like GitHub, the fake scopes each installation token to the repositories and permissions it was minted for. A Git write with the wrong token is refused, which backs the publication tests' token assertions. The service fork has its own installation (forkInstallation).
• publish.test.ts drops its test-local fake. It uses the shared one with a small world of its own, and replaces the fake's knobs (loseMove, onCreate, forkParent, head, role, now) with world edits and a shared intercept wrapper. A new test checks that the fake refuses out-of-scope writes.
• seed.test.ts adds an end-to-end run: a runner publishes the seeded queued job's commit, opens its draft pull request, and marks it ready. A retry of the failed import job then adopts the pull request it left.
• dev/serve.ts passes request headers to the fake, since token scoping needs them. The seeded server boots and serves /inbox.

Choices made without asking

• Token scoping in the fake. The old fake returned fixed fork-token/upstream-token strings. The shared fake mints distinct tokens and enforces their scope, so tests assert scope through the fake rather than through token names.
• Repo.branches replaces Repo.claims. It is a map of upstream branch to commit, so an iterated pull request's head branch can move. A quest/* entry without a pull request is still a claim. Issue.comments is now the comment list rather than a count.
• Seed consistency. The seed now records the fork branch quest/m0/import at pull request 20's head, so a retry of the failed import job can publish.
• Publication tests keep their own small world rather than reusing the rocket seed. Their numbers stay readable and the publish tests don't depend on seed data.

Validation

• just check and just test pass locally: 465 CLI tests and 335 Worker tests.
• just worker-seeded boots and serves the inbox.

Notes

• A sibling quest, quest/m0/worker-tests.md, also edits publish.test.ts (D1 state isolation) and will likely conflict here.
• quest/m0/kix/merged-claims.md plans to test against this fake. Its merged-head cases can now use branches and pull request heads.

Suggested follow-ups

• Other Worker tests (projects, inbox, jobs, push) still carry their own fetch mocks. Some might move onto the shared fake where they exercise flows it models. This is a candidate for /quest-plan, not done here.

🤖 Generated with Claude Code

(Written by Claude Opus 5.5)

[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 2026-10-24T15:55:20.753251Z	41fb6ad	New commits
🔒 Security Review	✅ Completed 2026-10-24T15:56:26.973685Z	41fb6ad	New commits

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.

[kixelated]

Landing summary

Landing head 41fb6ad with a squash merge.

Changes since the last reviewed head (8d95317)

• Merged main twice (no rebase) to pick up docs: land squash merges, and re-review any non-trivial change #206, Fix the runner's post-merge review findings #210, feat(kix): show what agents are working on as summary cards #215, Stop the lifecycle CI jobs failing at random #218, A branch named after an epic never blocks its claims #214, and feat(kix): plan from an idea, pasted context, or a job's follow-ups #205.
• Resolved the cloudflare/test/publish.test.ts conflict with Fix the runner's post-merge review findings #210's review-thread paging by keeping this PR's seeded GitHub fake. cloudflare/dev/github.ts now pages a pull request's review threads 100 at a time from the threads cursor, and Fix the runner's post-merge review findings #210's busy-pull-request test now runs against the seeded world.
• feat(kix): plan from an idea, pasted context, or a job's follow-ups #205's new plan test asserted on the removed inline fake's github.requests. It now reads the mocked fetch through requests(), like the rest of the suite. Git merged it without a conflict, but without this fix the test would have broken after landing.
• just check and just test pass locally. The required checks pass on 41fb6ad. Codex reacted with a thumbs-up to that head and raised no findings. quest check --base origin/main, run from a build of main, passes, and the PR deletes its finished quest quest/m0/kix/github-fake.md.

Decisions

• The maintainer chose to land without waiting for the optional Codex security review. It had been stuck on "running" for more than 8 hours, and Codex's code review of 8d95317 was clean.
• The optional native Windows (windows-11-arm) job fails in test/gate.test.ts ("concurrent processes never admit past the headroom"). This PR doesn't touch that test, and main fails the same job at 82a9566, so the failure doesn't block this merge.

(Written by Claude Opus 5.5)