[JacobStephens2]

Summary

0.9.0 lets a Command's sessions run on Codex as well as Claude Code. The headline is choosing the Harness, Model and Effort (thirdshift <Issue URL> harness codex model GPT-6.1-Sol effort Max), on the command line or in the User config: #401 (the Spec), #402, #404, #405 and #406.

• harness, model and effort on a Run, an Architect run and a Pickup run (Choose the Harness, Model and Effort on the command line and in the User config (#392) #402). The command line wins, then [harness], [harness.claude] and [harness.codex] in the User config, then the default: claude with the Harness's own Model and Effort.
  • The choice is checked before any work starts. That means the CLI is on PATH, a test call for a named Claude Model, and a match against codex debug models for Codex.
  • Every session gets the choice, including Repairs, Resumes, a Spec run's Tickets, the Spec review and a Base fix.
  • The Command log, the Activity log, the PR body (Built with claude · opus · high) and the Run notification all record it.
• Codex sessions (Run a Command's sessions on Codex #404). With harness codex, sessions run codex exec --json unsandboxed, as ADR 0012 records. Codex display names and Efforts are settled to Codex's own names, and progress lines, token counts and the Session log come from Codex's stream.
• Codex Resume and interrupts (Resume and interrupt Codex sessions (#395) #405). A command or sub-agent call left running gets the one Resume, as on Claude. An interrupt sends a Codex session SIGINT before SIGTERM.
• Setup asks for the Harness, Model and Effort (Setup asks for the Harness, Model and Effort (#393) #403, Setup offers Codex (#396) #406). A Claude Model is checked with a test call. For Codex, Setup lists the Models and Efforts from Codex's catalog and checks each answer the way a Run does. - clears a current Model or Effort. Re-running Setup adds the [harness] sections to an existing User config.
• Factory skills are linked into the worktree as thirdshift-<skill> instead of loaded as a Claude plugin (Load the Factory skills from the worktree as thirdshift-<skill>, replacing the Claude plugin #400, ADR 0012):
  • Claude finds them in .claude/skills/ and Codex in .agents/skills/.
  • They are kept out of git with .git/info/exclude.
  • Prompts now start /thirdshift-<skill> on Claude and $thirdshift-<skill> on Codex.
  • The missing resolving-merge-conflicts skill, which the conflict Repair prompts use, is restored.
• Labels (One label-edit module changes an issue's labels for the Claim, the Sweep and the Architect plan and idea #366). ready-for-agent and needs-triage are now created, with their descriptions, in a repository that lacks them. A Claim on an issue already in-progress that still carries ready-for-agent now removes that label, so the issue's other labels keep their order.
• An Architect plan is viewed once (Test the Architect plan's checks and the Architect idea's labelling in-process behind the Pass seam #387). Its label Edit is now built from that view, so a label added between the view and the Edit is overwritten.
• waiting on Base fix #n is now printed before the child Run starts, so it also shows when the start fails (Test the Base fix's rules in-process behind a seam for its outside effects #379).
• Internal refactors with no other change you can see. These move each part's rules behind a seam and test them in-process: the CI watch (The CI watch hands each Inherited failure its Base branch checks, and the Base fix reads none itself #364, The CI watch owns its rules, tested in-process behind a seam #399), the Spec PR and Spec run (One Spec PR module owns the Spec PR in a Spec run #369, Test the Spec run's Ticket loop in-process behind a seam (#380) #381, The Spec PR owns its rules, tested in-process behind a seam, and the Delivery writes "Built with" through its own #410, The Spec run owns its lifecycle to its Delivery, tested in-process behind its seam #420), the Ready issue search (Test the Ready issue rules in-process behind a seam for the search's reads #371), the Repair loop (Test the Repair loop's rules in-process behind a seam for its outside effects #373), the Passes (Test the Architect run's and Pickup run's skip rules in-process behind a Pass seam #375, A Pickup run owns its lifecycle to its dispatch, tested in-process behind the Pass seam #416, An Architect run owns its lifecycle to its dispatch, tested in-process behind the Pass seam (#415) #418, Each Pass owns its lifecycle, from its gates to the run it dispatches, tested in-process behind the Pass seam #417), Issue branch selection (Test Issue branch selection in-process behind a seam for its reads #377), the Base fix (Test the Base fix's rules in-process behind a seam for its outside effects #379), the Claim (The Claim owns its lifecycle, tested in-process behind a seam #383), the Delivery (The Delivery owns its steps and the Failed run path, tested in-process behind a seam #385), the Run's start (The Run owns its start, tested in-process behind a seam (#388) #389), the Sessions' Resume rules (The Sessions module owns its Resume rules, tested in-process behind a seam #408) and Setup (Setup owns its flow and questions, tested in-process behind a seam #412).

Also: docs for the Harness, ADR 0012 and the Codex research (#397).

Version diff

diff --git a/Cargo.lock b/Cargo.lock
index c8d7884..9bf359a 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1469,3 +1469,3 @@ dependencies = [
 name = "thirdshift"
-version = "0.8.4"
+version = "0.9.0"
 dependencies = [
diff --git a/Cargo.toml b/Cargo.toml
index 719640b..58d8fed 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -2,3 +2,3 @@
 name = "thirdshift"
-version = "0.8.4"
+version = "0.9.0"
 edition = "2024"