[JacobStephens2]

Summary

in-progress now marks only a live Run (ADR 0029). How a Run ends decides what replaces it:

 Claim::end(ending)
   Merged                      -> remove in-progress once the issue is closed
-  ReadyForReview | Held       -> keep in-progress
+  ReadyForReview | Held       -> remove in-progress, put nothing back
   Failed, nothing on origin   -> release (ready-for-agent back)
-  Failed, work on origin      -> add agent-failed beside in-progress
-  Security fix failed         -> add agent-failed beside in-progress
+  Failed, work on origin      -> swap in-progress for agent-failed (one edit)
+  Security fix failed         -> swap in-progress for agent-failed (one edit)
   Stop / NotReproduced        -> unchanged (needs-triage / needs-info)

Pickup's gates:

 gates
   sweep(closed issues: in-progress, agent-failed)
   clear_and_count_failures(open agent-failed issues)
-    neither in-progress nor needs-triage -> "took over": take agent-failed off, don't count
+    (no such take-over: a failure now leaves exactly that, so it counts)
     ready-for-agent back                 -> take agent-failed off, don't count
+    every one                            -> take in-progress off too (the Sweep), warning on failure

A Custom process already removed in-progress at ready for review, and it now shares the failure swap. Its Delivery now records whether its own pull request merged. A merged Custom process ends its Claim the way a Ticket run does: in-progress stays until the issue is closed. A child Run that the Process dispatched and that merged no longer counts as the parent's merge.

The guide and the hand-written pages (site/index.html, site/how-it-works/, site/architecture/) describe the new endings. site/docs/index.html is regenerated from the guide.

Evidence

Execution evidence

• Before (against a6880d0), for the Spec review's finding: cargo test --test process a_self_merge_whose_issue_could_not_be_closed_keeps_in_progress_until_it_is_closed

  test a_self_merge_whose_issue_could_not_be_closed_keeps_in_progress_until_it_is_closed ... FAILED
  

  After: cargo test --test process → test result: ok. 236 passed; 0 failed; 1 ignored

• Before (against f082b5c), for the Standards review's follow-up finding: cargo test --test process a_dispatched_child_that_merges_does_not_keep_the_parents_claim

  panicked at tests/process/published.rs:860:5:
  ["in-progress"]
  thirdshift: labelling #7 in-progress, in place of ready-for-agent
  thirdshift: labelling #8 in-progress
  thirdshift: removing in-progress from issue #8
  test result: FAILED. 0 passed; 1 failed
  

  After: cargo test --test process -- a_dispatched_child_that_merges_does_not_keep_the_parents_claim a_self_merge_whose_issue_could_not_be_closed self_merge → test result: ok. 14 passed; 0 failed

• These fake-gh integration tests pass on this branch:

  • pickup_run: 146
  • fresh_run: 26
  • spec_run: 92
  • merge_run: 40
  • continuation: 9
  • security_run: 125
  • architect_run: 94
  • base_fix: 16
  • stop: 10
  • bug_run: 14
  • fix_run: 11
  • triage_run: 28
  • activity_log: 16
  • review_pass: 14
  • failed_run: 15
  • process: 236

  pickup_run and process were rerun on the review fixes. The rest ran on the first commit; the later commits change only Custom-process code and wording, which those files don't assert. Locally I also ran cargo clippy --all-targets (clean), cargo fmt --check, and npm run docs:check --prefix site (passes).

• Not run locally: the bin's inline unit tests (claim::tests, pickup::tests). rustc was OOM-killed (SIGKILL) building the bin's test harness on this 7 GB host. They typecheck (cargo check --tests, clippy), and CI runs them. Two site checks also fail here for host reasons: npm test --prefix site has no Playwright Chromium, and site_deploy::the_systemd_units_verify trips on the host's own snapd.service. Neither touches the changed files.

Visual evidence

No headless browser is available on this host (Playwright's Chromium is not installed), so there are no captures. The site changes are text only: two sentences, one list item, and two SVG labels that now read "PR to draft, agent-failed" instead of "PR to draft, Claim kept". The SVG box widths are unchanged.

Unaddressed findings

Standards

• The Failed run path #5 — Mysterious Name: clear_and_count_failures now also does part of the Sweep. Declined. The function's main job is still counting failures, and its error handling differs from sweep's: a failed listing fails the run instead of warning. The two docs cross-reference each other (src/pickup.rs, the sweep doc: "[clear_and_count_failures] sweeps the open issues that carry agent-failed"), and the guide's Sweep section names the open-issue step. Renaming it would be churn with no change in behaviour.

Spec

None. The single Spec finding, a Custom process that merged without closing its issue, is fixed and tested.

Security review outcome

No unaddressed introduced findings.

Review coverage

Both reviewers read every file changed in a6880d0 and f082b5c. Both accepted site/docs/index.html as generated output and did not read it: it is generated from docs/guide.md by site/generate-docs.mjs, and npm run docs:check --prefix site passes. Commit a37987d came after the follow-up review, so the review left these unread:

• src/processes/ticket.rs: one merged: false initializer.
• tests/process/published.rs: the Standards reviewer's own test, renumbered to issues the fake has.
• The a37987d hunks in src/processes.rs, src/claim.rs, src/pickup.rs and docs/guide.md.

Merge Danger

Door: two-way

Only label edits change. Reverting brings back the old labels on new endings. Issues already relabelled stay valid under both rules.

Blast Radius: labels

• The first Pickup run after merge takes in-progress off every open issue that also carries agent-failed, on every repository it runs on. That is intended (ADR 0029), but it is a visible batch of label edits.
• Issues that ended ready for review under the old rules keep in-progress until someone removes it by hand. Spec: in-progress marks only a live Run #1901 leaves that out of scope.

Closes #1902

Built with claude 2.1.314 · claude-opus-5-5 · medium