diff --git a/.agents/skills/release/SKILL.md b/.agents/skills/release/SKILL.md
index 389cb44..baa926d 100644
--- a/.agents/skills/release/SKILL.md
+++ b/.agents/skills/release/SKILL.md
@@ -7,18 +7,18 @@ release the current changes. context (if any): $ARGUMENTS
 
 ## version locations
 - `build.zig.zon` line 3: `.version = "x.y.z"`
-- `src/relay/main.zig`: grep for `version=` in metrics endpoint
 - `CHANGELOG.md`: add new section at top
 
 ## steps
 
 1. **diff**: `git diff main --stat` + `git log main..HEAD --oneline` to understand what's shipping
 2. **decide bump**: patch (fixes/small), minor (new features/APIs), major (breaking changes)
-3. **update versions**: edit `build.zig.zon` and `src/relay/main.zig`
+3. **update version**: edit `build.zig.zon`
 4. **changelog**: add `## x.y.z` section at top of CHANGELOG.md
    - one line per change, prefixed with `**feat**:`, `**fix**:`, `**refactor**:`, `**docs**:`
    - match existing style (terse, technical, no fluff)
 5. **devlog**: if the work is substantial (new subsystem, major perf win, interesting technical story), suggest a devlog entry — ask before writing one
-6. **verify**: `zig fmt --check . && zig build test`
-7. **commit**: stage changed files, commit as `release: vx.y.z`
-8. **tag + push**: `git tag vx.y.z && git push origin main --tags`
+6. **bench**: `cd ~/tangled.org/zzstoatzz.io/atproto-bench && just bench-zig` — it pins zat by local path, so check out the previous tag, run, check out the candidate, run; interleave twice and compare within the run (30-40% run-to-run variance). a regression blocks the release
+7. **verify**: `zig fmt --check . && zig build test`
+8. **commit**: stage changed files, commit as `release: vx.y.z`
+9. **tag + push**: `git tag vx.y.z && git push origin main --tags`