run: with no model configured, each phase inherits Claude Code's user-level `model` and the run header says nothing about it

Open
  • Is something wrong?
#1,640 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

Nobody has claimed this yet.

Assessment

Difficulty
3/5
Estimated time
1-2 days
Newbie friendliness
42/100
Issue type
Feature
Clarity
Clearly specified
Activity status
Active
Tech stack
typescript
Domain
devtools

Research direction

Start with drivers/claude-code.ts around line 165, where the model is spread into the driver config only when config.model is set, then find where the run header is printed. The change is to print the resolved model per phase and label phases that inherit from Claude Code settings. Done when the header and summary show the model for each phase and a run with no model configured says it is inherited.

Written by the indexing model from the issue text.

Description

What happens

sequant run <issue> --phases spec,exec,qa with no --models, no modelLadder and no model in .sequant/settings.json passes no model to the driver. drivers/claude-code.ts:165 spreads ...(config.model ? { model: config.model } : {}), so the Claude Code process picks its own default, which is the model key in ~/.claude/settings.json. The run header prints Phases, Mode and Issues and no model. The only record is the per-phase model field in the run log (#1198).

Evidence (sequant 2.20.0 at 31c6f8bd, a 30-run batch on a private mirror, 2026-11-02 to 2026-11-03)

The owner's ~/.claude/settings.json changed at 00:12 CST on 2026-11-03 (file mtime), between two runs of the same batch. From the run logs' phases[].model:

  • #1143 F1–F5 and #1144 F1–F5 (started 16:50 to 22:26 CST on 11-02): every phase claude-opus-5-5.
  • #1153 F1–F5 (started 02:30 CST on 11-03 and later): every recorded phase claude-fable-5-1; in F3, F4 and F5 the exec field is null because the phase hit phaseTimeout and the driver reported no modelUsage, and the transcripts' model fields are claude-fable-5-1 throughout (305 to 593 messages per run).
  • The pilot the day before: #1143's F run on Fable, the other five on opus.

Nothing in the batch's stdout shows the change. The two groups differ in cost (opus runs $4.18 to $12.62; Fable runs $4.48 to $46.93 by the SDK estimate) and in how often they hit phaseTimeout (opus 0 of 10, Fable 3 of 5), so a comparison across them is a comparison of models, which the run output does not say.

Ask

  1. Print the resolved model for each phase in the run header (Models spec=… exec=… qa=…), and when a phase has no configured model, say so: exec: inherited from Claude Code settings (claude-fable-5-1).
  2. Put the same line in the summary table or the final Log: block, so a log read after the fact carries it without opening the JSON.
  3. Optional: a run.requireModel setting (default off) that refuses to start a phase with no configured model, for benchmark and replay use.

Why it matters

A run's model is the largest single cost and behaviour variable, and today it can change between two issues of one batch with no trace outside the run log. #1198 made the record exist; this is about making it visible at the time.

From a private replay study of sequant on our own repos; raw runs are not public. Filed by the owner.

Dominant language
TypeScript
Stars
1
Forks
1
Avg merge
11h 3m
Merged PRs (30d)
84

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from sequant-io/sequant

All issues in sequant-io/sequant

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.