# Release notes Release notes are published in English with every release. ## 1.1.0 ### Minor Changes - 16d7416: Close the capability gaps against the upstream kit, clear the inherited reference debt, and make the eval coverage claim true. **Two new skills.** `av:av` documents the `av` CLI itself — nothing in the kit did, so an agent operating this control plane had to guess. It is written from live `--help` output across every command and points at `av --help` as the authority rather than duplicating flag tables. `av:plan-i18n` adds the bilingual Vietnamese/English switch for `plan.html`, re-scoped to that artifact alone and deferring the planning workflow to `av:plan`; upstream's instructions for subcommands `av` does not have were deleted rather than guessed at. **`av update --to `.** Update could only move forward to latest, so a user who hit a regression had no way back except re-running the installer by hand. `--to` installs one exact release through the edge's pinned selectors, with the version validated before it reaches a URL and checksum verification mandatory — a downgrade is exactly when a bad binary is hardest to spot. **`av validate --strict`.** 89 reference files were on disk that no `SKILL.md` mentioned; most were navigation lists written with bare filenames the integrity checker never matched. Each was read and then linked, indexed with a stated purpose, or deleted. With the backlog at zero, `--strict` promotes orphan and dangling findings to errors and CI runs it. **Eval coverage that enforces itself.** The suite documented full skill coverage while 77 of 103 skills had no scenario. There are now 105, authored by confusable cluster so that a negative case names a skill a model would plausibly have picked, and the coverage tests read `kit/skills/` at runtime so the claim cannot drift again. The evidence vocabulary grows from 27 ids to 40. No behavior changes for existing installs beyond the new skills and the new flags; `av update` with no flags is unchanged.