feat(cli): add pull request checkout command

[yan-ad]

Summary

• add crabcode pr <number> to fetch and checkout GitHub pull requests as pr/<number>
• configure fork remotes and upstream tracking for cross-repository pull requests
• relaunch Crabcode after checkout and document the command

Validation

• cargo fmt --check
• cargo check
• focused CLI and PR module tests
• crabcode pr --help smoke test

Notes

The full suite currently has 6 unrelated failures in discovery, compaction, jobs, and UI; 1408 tests pass.

---

Review: yan-ad:feat/pr-command → Blankeos:main

Merge confidence: 4.5/5 — ready to merge based on targeted validation.

Reviewed candidate: 6345aae, including follow-up fixes in 6611919. Small feature scope; no migrations required. The reproduced checkout safety, missing fork ref, and repeat-launch problems are fixed. Remote alias/collision handling is now covered too.

Changes pushed

• Removed unconditional forced checkout, retaining GitHub CLI safeguards for local edits and commits.
• Fetch the fork branch before configuring its upstream.
• Before adding a fork remote, preserve the already-resolved base repository with gh repo set-default. This prevents adding the fork from breaking subsequent interactive checkout through repository ambiguity.
• Match fork remotes by repository identity, reusing existing aliases. If an owner-named remote points elsewhere, choose a nonconflicting name without changing the unrelated remote.
• Added targeted regression coverage for safe arguments, missing refs, failed fetch/pin behavior, alias reuse, host identity, collision preservation, and pin-before-add ordering. No new dependencies.

Checks performed

• cargo check --locked and cargo build --locked passed.
• cargo fmt --check and git diff --check passed, including pre-commit validation.
• 12 focused tests passed: 10 PR-module tests, CLI parsing, and root help.
• Earlier binary crabcode pr --help smoke check passed.
• Newly built candidate tested through a controlling pseudo-terminal in disposable /tmp repositories, without model prompts:
  • Fresh-fork PR feat(cli): add pull request checkout command #50: checkout, correct upstream, home-screen rendering, and /exit all passed.
  • Repeat PR feat(cli): add pull request checkout command #50: launch and clean exit passed without manual default-repository setup; staged and unstaged edits were preserved.
  • Same-repository PR feat(update): lazy version check with clickable upgrade toast #54: checkout, home-screen rendering, and clean exit passed.
• Earlier real-GitHub repeated-checkout check confirmed local commit HEAD preservation with the non-forced checkout arguments, which remain unchanged.

Remaining limits / regressions

• No regression identified in the reviewed command paths. Full-suite results do not establish a clean base/candidate comparison; baseline classification remains inconclusive. The original Notes section's six failures/1,408 passes are historical author-reported results, not current independently verified totals.
• Repositories already made ambiguous by older builds may still need a one-time gh repo set-default OWNER/REPO. The new code prevents creating that ambiguity; it deliberately does not guess the intended base in an already-ambiguous repository.
• End-to-end smoke checks were on macOS with GitHub CLI 2.100.0; no cross-platform runtime guarantee.

Migrations

None. No database, configuration-file schema, credential, or dependency-format migration. The command intentionally changes checkout/remotes/upstream settings; the follow-up additionally records the confirmed GitHub default repository in local Git configuration before adding a remote. Git and authenticated gh remain required.

Merge checklist / confidence bumps

• Fix the reproduced blockers and cover remote aliases/collisions.
• Pass build, formatting, focused tests, and fresh/repeat/same-repo interactive launch checks: 4.5/5.
• For 5/5 confidence: establish a repeatable full-suite base/candidate comparison with no unexplained new failures; supported-platform smoke checks provide further assurance.

No additional manual launch test is needed to reproduce the evidence above. The remaining full-suite comparison is extra assurance, not an identified blocker. Confidence scores are judgments, not guarantees.

[Blankeos]

Review: yan-ad:feat/pr-command → Blankeos:main

Merge confidence: 4/5 — small, isolated change; the two reproduced blockers are fixed in 6611919. One interactive setup caveat remains.

Follow-up changes

• Removed unconditional gh pr checkout --force, preserving GitHub CLI's safeguards for local commits and uncommitted edits.
• Fetch the fork's head branch into its remote-tracking ref before assigning upstream, fixing first-time fork checkout failures.
• Added three regression tests: safe checkout arguments, fetching a missing fork ref, and preserving upstream configuration when fetch fails. No new dependencies.

Verified checks

• Formatting and diff whitespace checks passed.
• cargo check --locked, cargo build --locked, and crabcode pr --help passed during validation.
• Focused tests rerun before committing: 7 passed (5 PR-module tests, CLI parsing, root help).
• Real GitHub checkout in disposable repositories preserved staged edits, unstaged edits, and local commits.
• Installed binary tested through a controlling pseudo-terminal in /tmp, with no model prompts: fresh-fork PR feat(cli): add pull request checkout command #50 rendered the home screen and exited via /exit; same-repository PR feat(update): lazy version check with clickable upgrade toast #54 did likewise.
• Repeat PR feat(cli): add pull request checkout command #50 launch preserved staged/unstaged edits and exited cleanly after setting the default GitHub repository.

Remaining caveats / regressions

• Interactive repeat-checkout caveat: adding the fork remote leaves two remotes. In a fresh repository without a gh default, reopening failed with “No default remote repository has been set.” Running gh repo set-default Blankeos/crabcode in that disposable repository resolved it. This remains a usability issue, not data loss; do not treat repeat-launch validation as unconditional.
• Non-blocking edge case: an existing contributor-named remote is trusted without verifying its URL. Remote identity/alias handling is recommended follow-up hardening.
• No existing-app regression identified in this isolated diff, but the full suite was not independently rerun or compared with base. The six baseline failures mentioned above are author-reported, not independently confirmed by this review.

Migrations

None. No database, config, credential, or dependency format migration. Git and authenticated gh are required; checkout and remote/upstream configuration are intentional Git side effects.

Before merge / confidence bumps

• Fix both reproduced blockers and pass focused regression checks: 4/5.
• Resolve or explicitly document the default-repository requirement and verify a fresh fork can be reopened without an unexplained setup failure: 4.5/5. Other interactive launch smoke tests already passed.
• Additionally cover remote identity/alias collisions and compare full-suite results against base with no unexplained new failures: 5/5.

Scores describe remaining risk and evidence, not a guarantee. The fixes above are pushed; review/manual instruction files were not included in the code commit.

[Blankeos]

lgtm