Skip to content

PDD 0.1.27

Choose a tag to compare

@Serhan-Asad Serhan-Asad released this 10 Nov 11:48
9341f3b

PDD 0.1.27 release notes

Release: https://github.com/promptdriven/pdd/releases/tag/alpha-v0.1.27
Release PR: #3090.

Summary

Autopilot reliability release on top of 0.1.26 (main). It collects the work that ran on pdd-vm as internal builds 0.0.1 to 0.0.14 (Nov 8 to Nov 10).

Install or update

npm install -g github:promptdriven/pdd
npx skills add promptdriven/pdd -g
pdd --version   # prints 0.1.27

Update the command and the skills together, then start a new agent chat. Running Autopilot? Update every machine on a repository, then run pdd autopilot start again.

What changes for you

  • Runs continue after a restart. After a machine or Herdr restart, a run opens its next chat with the same identity and evidence (#3084, #3113, #3123).
  • PR 1 can change after code. A bug or feature run can go back to the prompt step after code work, update PR 1 and check it again. Its claim and valid approvals stay (#2897, #2894).
  • Honest PR evidence. The PR shows only the Stories that the run changed, and the evidence that was really recorded (#3085, #3088, #3114).
  • pdd autopilot doctor. It checks setup and machine readiness and prints a fix for each problem (#3125).
  • Post-merge audit. Autopilot audits each merge and reports when a change may not fully fix its issue (#3111, #3138).
  • Intake does not stop on one failure. Assigned issues are skipped. One failed sort, split or route keeps its own failure, and other issues continue.
  • Free slots are used. Completed and cancelled runs no longer hold a run slot (this held all new PDD starts on Nov 9 and 10). A failed wake of a parked run no longer stops unrelated ready issues (admission fix).
  • Cancel and claim cleanup. pdd workflow cancel stops PR publication, and it releases stale claims and labels after cleanup (#3157).
  • No local interpreter queue. The client sends checks directly again. Recovery of the same request and the person-approved resend of a lost result stay.
  • Tests leave your machine alone. The timer replay test no longer creates real launchd jobs on macOS.
  • Small changes check every requested item (#3139). Code-only bug fixes have their own checker route (#3131).

Known limits

  • Some bug and feature runs still stop when PDD cannot check a required route or interface. The interface-port, routing, Runtime-wake and self-report work is not in this release.
  • GitHub's ownership reads and label deletes are not one atomic step, so a claim release can leave the local cache unchanged until the next tick.

Proof on the exact head (52d9b10)

  • Source reviews: Codex and Gemini final integration reviews, no remaining blocker.
  • Prompt approvals: 7 Runtime contracts recorded with pdd approve --said; 19 prose-only decisions recorded by the lead.
  • Full VM suite on 52d9b10: 8,387 of 8,388 passed, 0 failed, 1 skipped (commit status vm-suite).
  • CI tests (25 min), install-smoke, CodeQL and Analyze: green on 52d9b10. The CI test job limit is now 60 minutes, because the suite has about 8,400 tests.
  • Live trial on pdd-vm (internal build 0.0.14 = same code, installed Nov 10 10:51 UTC): PDD issue #2830 started 20 minutes after the install; both Autopilots stayed healthy; no stop of all runs. On the older build, ended runs kept their run slots and blocked new PDD starts until the lead removed their stale files by hand.

Diff: 297 files changed, 24,343 lines added, 2,606 lines removed.