[JacobStephens2]

Summary

A Spec whose Tickets are all closed, with nothing started, is no longer a Ready issue: a Pickup run passes it over with its own line and goes on to the next candidate.

 standing_of(candidate)            # the Ready issue rule, first reason wins
   Unready label        -> "#<n> labelled <label>"
   Claim                -> "#<n> labelled in-progress"
   sub-issue            -> "#<n> is a Ticket of #<spec>, which is not ready"
   base-fix             -> "#<n> labelled base-fix"
   open blocker         -> "#<n> blocked by #<m>"
   already started      -> "#<n> already started: ..."
+  every Ticket closed  -> "#<n> every Ticket is closed"
   not settled          -> "#<n> not settled: ..."
   otherwise            -> Ready issue

The candidate query now carries each sub-issue's state, so no extra request is made per candidate, and both the Pickup run and the Spec run's own check go through spec_run::all_closed:

 CANDIDATE_QUERY
   parent { url }
-  subIssues { totalCount }
+  subIssues(first: 100) { nodes { state } }

thirdshift <Issue URL> on such a Spec is unchanged: it still stops with every Ticket is closed and there is no Spec branch; nothing to do. CONTEXT.md, the README's Ready issue rule and "Why an issue was passed over", and the pickup help text describe the new condition.

Evidence

• Before: the four new tests and the extended first-reason test in tests/pickup_run.rs failed against main's code:

  with_only_a_spec_whose_tickets_are_all_closed_the_pass_is_skipped_changing_nothing
    stderr: thirdshift: taking Ready issue #7 "Issue 7", as thirdshift https://github.com/acme/widgets/issues/7 would
            thirdshift: every Ticket is closed and there is no Spec branch; nothing to do
    left: Some(1)   right: Some(0)
  a_spec_whose_tickets_are_all_closed_is_passed_over_and_the_next_ready_issue_is_taken
    no PR from issue-9: []
  a_spec_whose_tickets_are_all_closed_is_never_said_to_be_not_settled
    left:  "thirdshift: #7 not settled: labelled ready-for-agent less than 10 minutes ago\n..."
    right: "thirdshift: #7 every Ticket is closed\n..."
  

  After: cargo test --test pickup_run: 73 passed, 0 failed. They cover the passed-over line, the next Ready issue being taken, the skip (exit 0, GitHub state unchanged, no Run notification), the already started line kept for a Spec branch on origin or a pull request, a Spec with one open Ticket still taken, and every Ticket is closed ahead of not settled.
• The full suite (cargo test, all 30 test binaries) passed on the first commit, with cargo fmt --check and cargo clippy --all-targets clean. After the review fixes, a rename and a doc comment, pickup_run, spec_run, preflight and fakes were re-run and pass; the full suite was not run again.

Merge Danger

Door: two-way

Reverting restores the old rule; nothing on GitHub is written differently.

Blast Radius: Pickup

Only what a Pickup run takes changes. A Spec with more than a hundred sub-issues is judged on its first hundred, as a Spec run already judges it.

Unaddressed findings

Standards

• The "nothing started" test differs between the Spec run's check and the Pickup run's (Selection::Fresh in src/run.rs, branch::started in src/pickup.rs): both predicates were there before this change, and the issue fixes the new reason's place as after "already started", so they were left as they are.
• Possible Speculative Generality: all_closed takes an iterator of open flags, not &[Ticket]: kept, since it is what lets one rule serve the Tickets read and the candidate read, which the issue asks to agree.
• Possible Primitive Obsession: Vec<bool> for a candidate's sub-issues: partly addressed (renamed sub_issue_is_open, with has_sub_issues() hiding the empty case); a type of its own seemed more than one caller needs.
• Possible Shotgun Surgery: the Ready issue rule is restated in CONTEXT.md, the README, the help, a doc comment and two tests: this is how the repository already documents the rule; gathering it is a separate change.
• Possible Duplicated Code: "every Ticket is closed" appears in the Pickup run's line and the Spec run's error: two different messages the issue asks for word for word, each used once.

Spec

• The first hundred sub-issues only: a Spec with more than a hundred Tickets whose first hundred are closed is passed over even if a later one is open. The Spec run reads the same hundred and would refuse it, so the two agree; paging is out of this issue's scope. The limit is now said in the query's doc comment.
• Not asked for: the pickup help text and its test: kept, because the help restates the Ready issue rule and would otherwise be wrong.
• Not asked for: the README's advice to close such a Spec or take its ready-for-agent off, and the example's pull request renumbered to free #33: kept, the first as the way out for the reader, the second so the example has no issue and pull request sharing a number.

Closes #303

🤖 Generated with Claude Code