[pasrom]

Release km 1.10.0. Merging this PR is the release (CI tags v1.10.0 once main is green). It ships the four changes merged since v1.9.0 (the retired paragraph check, workflow_dispatch in the team ci workflow, the bump-mounted-pins workflow and script, Markdown one paragraph per line) and three commits here:

1. docs(skill): skills/km/reference/setup.md names main-guard and bump-mounted-pins (with its script) among what km init --team writes, adds a short section "Mounted pins on a team brain" (what it does; the maintainer's setup before the first mount, taken from the workflow header: GitHub App with Contents and Actions read, environment mounted-pins restricted to main created before the first run, secrets CI_APP_CLIENT_ID and CI_APP_PRIVATE_KEY, Actions allowed to create pull requests, "Require approval of the most recent reviewable push", optional MOUNTED_PINS_RUNNER; the header stays the source for details), says what km upgrade adds to a team brain, and that a doctor note whose fix is in the repository settings is the maintainer's. No sentence in skills/ or the README said Dependabot moves the mounted pins; the Dependabot lines (km pin, action pins) stay, one now adds "or another action".
2. test(skill): line limits replaced. The core's 500-line limit goes: at the hard-wrapped 76 bytes per line it was about 38,000 bytes, three times CORE_BYTES (12,000, kept), so it never bound. The contents-list rule becomes CONTENTS_BYTES = 6000: the old 100-line rule split the reference files at 3,433 to 3,702 bytes (under) against 8,064 to 13,599 (over), and 6,000 keeps that split. New: no line in a Markdown file under skills/ over 1,900 characters (the Read tool cuts at 2,000); longest today is 1,875 in queue.md (Open PRs step), so the next growth there means splitting it. skill_smoke.sh mutation-checks each new check (a padded core, brains.md without its contents list, a 1,901-character line fail; a 1,000-character line of two-byte characters passes), and disabling each check or counting bytes fails the smoke test.
3. chore(release): src/km/__init__.py, .claude-plugin/plugin.json, README examples (pyproject reads the version from __init__; no CHANGELOG in this repo).

Evals

claude plugin eval . --trust-plugin --case doctor --scaffold --no-publish --ablation none --allow-tools Bash Write Edit (setup.md maps to the doctor row; no case covers init or upgrade): doctor 1.00, 3/3 runs, both graders (reads-reference, runs-km-doctor) passed every run, $0.83, 131 s. Caveat: the run started on the first version of the setup.md section; the simplify and review folds shortened that section while it ran, so not every run read the final text. No grader changed, so no baseline.

Review

/simplify (reuse, simplification, efficiency, altitude) applied: the new setup.md section no longer copies the five setup steps verbatim (one sentence naming the items, the header stays the source), the repeated "before the first mount" warnings in init and upgrade became pointers, the doctor rule is general (a note fixed in repository settings), text bound once in the inventory loop, one printf idiom for filler. Skipped: trimming the calibration comment (the numbers are the point of the change); reading the skill files once instead of twice in the line check (negligible, and the scan covers all of skills/); also scanning src/km/templates/**/*.md or adding a line check to km validate (outside this PR; templates are under 1,500 characters today), worth a follow-up.

/code-review applied: the CORE_BYTES mutation pads 12,001 bytes so it cannot go stale when the core shrinks; the section no longer says the workflow "does nothing" on a brain with mounts that is not set up; lines are split at \n only and read as bytes (universal newlines would split at a lone \r); ## Contents must be a heading line; the sync sentence says "usually" the workflow's PR (a hand-made pin PR stays fine). Skipped: UTF-16 counting of astral characters (the skill's text is BMP; the 100-character margin covers ordinary use); --skill pointing elsewhere widens the rglob (default and smoke point at skills/km).

All smoke tests pass locally.

Rerun on the final head 023dc5a: doctor 1.00, 3 of 3 runs, $0.78.