release: 0.13.0
Release commit for 0.13.0: moves the 27 unreleased commits' changelog entries into ## 0.13.0 - 2026-10-16, leaves an empty ## Unreleased, and sets package.json to 0.13.0. This pull request does not tag, publish, create a GitHub release or touch the npm publisher.
Why 0.13.0. The package is 0.x, so a minor bump marks the incompatible changes. Two interfaces changed: benefit-chart.py now requires band in a saved spec (it defaulted to 0.03 before), and checkBoundary accepts only version-2 handback records. The release also adds the web-image-formats modifier, the hr-handoff successor reference and several hr-benchmarking additions. A patch number would hide the two breaking entries. The two breaking entries open the section and stay whole.
Unreleased commits: 27 past v0.12.0 (0fd336b), counted with git rev-list --count v0.12.0..main.
Changelog audit (names, commands and counts checked against the code that ships):
benefit-chart.py:bandis required and read inband_of; a change at or inside it is labeled "no clear change". Supported.scripts/continuity-boundary.mjsis repository-only (thefileslist excludesscripts/), and the entry says so. Version 2, thecontextfields, the receipt match and the exit codes were read from the source.skills/hr-what-next/scripts/check-open-work.mjs(shipped) takes--sessionand--turntogether and not with--archive.oha --latency-correctionneeds-q:skills/hr-benchmarking/references/tools.mdline 13.- Run-count rule (3 plus 2, cap 5) and the audit exception (4 runs, 9 per arm for a smaller gap) match
hr-benchmarkingandprune-workflow.md. - Muse Wonder Full figures (xhigh 41.5, max 42, two runs each, no list price) match
panel-recommendations.json. - Iteration entries for the band removal, run counts and benchmarking guidance were merged into one entry per concern. No transient numbers remain.
Public-materials audit (existing evidence only, no new benchmarks):
| Material | Result |
|---|---|
| README modifier table | 20 modifier files, each named in the README, including web-image-formats. |
| README skills and eval claims | 39 skill directories, each named. The eval claim (18 scenarios, 2026-10-11, revision efd4d7b) is labeled historical; evals/end-of-reply/scenarios still holds 18. The new decision scenario is under evals/skill-fires, so those counts stay true. |
| Overview graphic and alt text | Describes layers, eight skills, no modifiers; unchanged by this release. |
package.json description | Still accurate. |
| Stale version numbers | No 0.11.0 or 0.12.0 in the README, docs or package.json. |
| 3% band leftovers | None in shipped guidance. band: 0.03 remains only as an example value in the benefit-chart.py docstring and as explicit test inputs, and one eval fixture states a 3% noise band as its own scenario input. |
| Private names and paths | None found in README, docs, rules, skills or evals. |
| Package contents | npm pack --dry-run lists 236 files, matching check:package. |
Accuracy blockers: none. One caveat for the owner. The v0.12.0 tag still ships the 3% default band that this release removes. That affects the failed publish run 38615008492 and any republish of that tag. That is expected, not a regression, and the 0.13.0 notes say the band is gone. npm still has 0.11.0 until a publish succeeds.
The separate rule pull request (#329) adds an Unreleased entry; land this one first so that entry stays under Unreleased.
Summary by cubic
Prepares the 0.13.0 release: moves the 27 unreleased commits' changelog entries into ## 0.13.0 - 2026-10-16, leaves an empty ## Unreleased, and bumps package.json to 0.13.0. This PR does not tag, publish, create a GitHub release, or touch the npm publisher.
Breaking changes
The 0.x minor bump marks two incompatible changes:
benefit-chart.pynow requires the spec's top-levelband, which used to default to 0.03.checkBoundaryaccepts only version-2 handback records; version-1 records and two-argument calls fail, unless the host supplies a stop, hold, pause, or low-quota signal that bypasses validation.
The release also adds the web-image-formats modifier and several hr-benchmarking additions. The hr-handoff successor reference clarifies that items the host cannot bind stay with the outgoing thread, and session replacement is confined to stale injected guidance that is session-critical. The variation criterion is defined as the comparison's own run-to-run spread. Land this PR before the separate rule PR (#329) so its Unreleased entry stays under Unreleased.
Summary by CodeRabbit
- Documentation
- Added the 0.13.0 changelog, including breaking changes to saved benefit-chart specifications and reply-boundary host requirements.
- Expanded decision-question guidance, documented the opt-in
web-image-formatsmodifier, and added successor-thread handoff guidance. - Updated benchmarking guidance for variation criteria, run counts, candidate versions, component and integration comparisons, performance investigations, and profiling.
- Added a Muse Wonder Full 1.0 panel recommendation and rechecked existing model figures.