[ethan-scitix]

Type

• chore — CI, tooling, dependencies, config

Summary

• Release 0.13.1: CHANGELOG entry and Dockerfile wheel version, as in the 0.13.0 release (chore: release 0.13.0 #237).
• A patch release: a max_tokens-derived context need reserves room for the prompt (fix(cli): reserve prompt headroom in a max_tokens-derived context need #239) without moving a task that fits under the native window into a scaled group (fix(cli): serve a max_tokens below native at the native window #240).
• Stacked on fix(cli): serve a max_tokens below native at the native window #240. This branch contains fix(cli): serve a max_tokens below native at the native window #240's commit so CI runs both together. Merge fix(cli): serve a max_tokens below native at the native window #240 first. The diff here then shrinks to CHANGELOG.md and Dockerfile; if it does not, rebase onto main.
• The migration note was checked against both versions' plan_serving_groups (native 40960, recipe 32768). 0.13.0 served max_tokens 38000 at 38000, 32768 and 30000 at 32768, and 60000 in m@60000. 0.13.1 serves the first three at 40960 and 60000 in m@64096. A min_context-only task keeps its window. The served window is part of the inference identity, so a changed window refuses the resume.

Related Issues

Refs #238

Test Plan

Automated

• Lint/format clean (ruff check . && ruff format --check .)
• Type check clean (ty check)
• Unit tests pass: pytest tests/unit tests/integration tests/acceptance -m "not benchmark" -n auto: 9692 passed, 94 skipped. One failure, tests/unit/vendor/code_evaluator/test_exec_quotebench_anchor.py, also fails on main on a macOS host (it compares against GNU verdicts).
• Acceptance gate pytest tests/acceptance -s: 12 passed, 10 skipped (staged BFCL archives and the code-evaluator service are not available locally)
• Benchmark gate pytest -m benchmark -s: all scenarios PASS (single-turn 576 SPS / 90.0%, long output 158.7 / 74.4%, multi-iteration 181.8 / 85.2%, high concurrency 1087 / 84.9%, framework overhead 1676.5 SPS, resume 0.5 s, peak RSS 408.7 MB)
• check_preflight.py --level deep: all pass except check_version (latest tag 0.13.0 vs CHANGELOG 0.13.1, which resolves when v0.13.1 is tagged after merge) and link warnings for placeholder, API-root and not-yet-existing compare URLs

Checklist

Required (all PRs)

• PR title follows conventional format (type(scope): description)
• No internal paths, credentials, or personal info in committed files
• AI-generated code has AI-Generated Code - <model> (<provider>) in module docstring
• No new upper-layer dependencies added to core/
• Deleted code verified — no remaining call sites depend on it

🤖 Generated with Claude Code