create-update-benchmarklisted

# Create / Update a Benchmark `.odd/benchmarks/<name>/` is a **third kind** of `.odd/` content, and it does **not** inherit the report stores' immutability rule (`create-observe-run-report`, `create-otel-instrumentation-report`): `AGENTS.md`'s "the `.odd/` memory is append-only" and `docs/guide/reports.md`'s "a report is never edited after the fact" govern the **committed reports** specifically - `observe-run-reports/`, `otel-instrumentation-reports/`, and `decisions.md`. A benchmark is living source, not a run record: git history, not file accumulation, is its memory. Writing and updating is this skill's whole point, not an exception to some other rule. ## What this skill owns - **Persisting** the k6 script and manifest `k6-benchmark-expert` hands it, into `.odd/benchmarks/<name>/` - the directory name is the benchmark's identity. - **Recall, two-step**: the target **service** returns the set of benchmarks that already exist for it (so the agent cannot duplicate one it never saw); the benchmark **name** identifies the single artifact an update rewrites. List every benchmark under `.odd/benchmarks/` and check each manifest's declared target service before the agent authors anything new. - **Reviewed diffs, never silent overwrites.** When the agent proposes updating an existing benchmark, the change is presented as a diff against the stored version - the maintainer reviews it exactly like any other committed change, through the normal PR flow. This skill