Introduction
dev-orchestra is a plugin for Claude Code / Codex / Antigravity that takes a one-sentence request through design, implementation, testing, review by several AIs, and fixes.
This post rounds up the changes in v1.3.2 (2026-11-09). The changes in v1.3.1 are in the previous release post. v1.3.2 is a bug-fix release with no new features.
dev-orchestra is open source on GitHub (MIT license). If it looks useful, a star would mean a lot.
Main changes
A resumed design run records only its own cost
When a design has to be revised, dev-orchestra can resume the previous design session and have it continue from there (run architect --resume).
Until now, when the architect was Claude, a resumed run was recorded with a cost that included the whole session it resumed. That is because the Claude CLI reports a resumed session’s cost (total_cost_usd) as the total for the whole chain so far. Token counts, on the other hand, come back for that run alone.
So each revision added the earlier runs’ cost again, and the architect’s cost came out higher than it really was in these places:
cost_usdon therunend eventarchitectandarchitect:resumedintokens show- the resumed-run ratios in
optimization report
In 1.3.2, the parent session’s cost is subtracted, and what is left is recorded as that run’s cost. The total the CLI reported is kept on the end event as resume.session_cost_usd, and the next resumed run subtracts it. When the figure to subtract is not known (for example, the parent run did not record a cost), the cost is recorded as unknown (null) and stderr shows note: this continued run's own cost is unknown; recorded without one.
Token counts are recorded as before. Also, events recorded before you update to 1.3.2 keep their figures, so read records from before the update apart from those after it. Only Claude architect runs have the parent’s cost subtracted; records for a Codex architect are unchanged.
Comparing 1.3.1 and 1.3.2
In a scratch repository, I ran the same design request once, resumed it twice, and compared the recorded costs. The architect was Claude’s sonnet, on Claude Code 2.1.322.
First, 1.3.1. The three runs were recorded as $0.0885 in total.
$ dev-orchestra tokens show
stage meas. input output total billed cost tools tool out
architect 3/3 14 1,245 - 5,786 $0.0885 5 345
ALL 3/3 14 1,245 - 5,786 $0.0885 5 345
Per reviewer and tier:
architect:resumed 2/2 4 470 - 1,787 $0.0645 0 0
The cost_usd on each run’s end event grows: $0.0239219, $0.0291082, $0.0354341. The second and third are not that run’s cost but the running total. What was actually spent is only the third run’s total, $0.0354341, so the record was about 2.5 times the real cost.
Next, 1.3.2. With the same steps, the three runs were recorded as $0.0336 in total.
$ dev-orchestra tokens show
stage meas. input output total billed cost tools tool out
architect 3/3 12 1,180 - 5,617 $0.0336 4 331
ALL 3/3 12 1,180 - 5,617 $0.0336 4 331
Per reviewer and tier:
architect:resumed 2/2 4 509 - 1,874 $0.0121 0 0
On the third run’s end event, cost_usd is that run’s cost and resume.session_cost_usd is the total the CLI reported (an excerpt of the event):
{
"stage": "architect",
"status": "ok",
"cost_usd": 0.006636,
"resume": {
"requested": true,
"mode": "resumed",
"outcome": "ok",
"trust": "newer",
"session_cost_usd": 0.0335753
}
}
The second run’s total was $0.0269393, and $0.0335753 − $0.0269393 = $0.006636 is the third run’s cost. Adding up the runs ($0.021508, $0.0054313, $0.006636) gives $0.0335753, the third run’s total.
Resuming a chain recorded by 1.3.1 after the update also records only that run’s cost. When I resumed the 1.3.1 chain above once more with 1.3.2, the CLI reported a total of $0.0420237; subtracting the $0.0354341 total that 1.3.1 recorded for the third run left $0.0065896 as cost_usd. The figures 1.3.1 recorded stay as running totals.
Smaller changes and bug fixes
- Claude Code 2.1.319 and codex-cli 0.166.0 were added to the CLI versions checked for resuming. Newer versions (with the same major version) are trusted to resume as newer than a checked one. With 2.1.322, each resume printed
note: claude 2.1.322 (Claude Code) is trusted to resume as newer than 2.1.319 (Claude Code), … - In the CHANGELOG entry for 1.3.1, the direction of the cost change from narrowing the tools passed to write runs was backwards; it is fixed (narrowing makes it 26% cheaper, from $1.72 to $1.28)
Stabilization
I added tests that record the cost of resumed design runs and did other work to keep things stable.
How to update
How to update depends on the CLI you use. Follow the steps in How to Update with the new version’s tag (v1.3.2). After updating, running dev-orchestra doctor once is a good idea.
Summary
- A resumed design run now records only its own cost. With a Claude architect,
run architect --resumeused to record the parent session’s cost as well; from 1.3.2 the parent’s cost is subtracted, and the total the CLI reported is kept inresume.session_cost_usd - In my test, 1.3.1 recorded about 2.5 times the real cost. In 1.3.2, the runs’ costs add up to what was actually spent
- Figures recorded before the update are not corrected. Token counts were already right
GitHub: istb16/dev-orchestra (a star would be a big encouragement)
Related Posts
[Released] dev-orchestra: A Pre-Design Interview, and Less Work in Reviews v1.4.0
A roundup of dev-orchestra v1.4.0. It adds an interview that asks one question at a time before the design (design.ask_questions: grill). In reviews, reviewers are told what was already decided, minor pre-existing problems stay out of triage, and re-review rounds run fewer reviewers. The CLI also picks the light tier itself, and a workflow can continue in a new session.
[Released] dev-orchestra: Start the Review While the Tests Run v1.3.1
A roundup of dev-orchestra v1.3.1. The code review can now start without waiting for the tests to finish (off by default). Plan approval now defaults to auto, which skips the approval when nobody is watching, such as in CI, and the conditions are met. Claude write runs now start with only the tools they use and no MCP servers.
[Released] dev-orchestra: A Closing Review of the Whole Change v1.3.0
A roundup of dev-orchestra v1.3.0. When a review round that looked only at the fix finds nothing above Low, the review now ends with one more round over the whole change. It is on by default, and a Medium finding now calls for another round, so a review can take more rounds than before.