astroclaw-test-performance

Files (2)

• SKILL.md
• openai.yaml · agents/

SKILL.md (first lines)

# Astroclaw Test Performance

Use evidence first. The goal is real `pnpm test`, plugin-suite, and
plugin-inspector speed/RSS improvement with coverage intact, not runner tuning by
guesswork.

## Workflow

1. Read the relevant local `AGENTS.md` files before editing:
   - `src/agents/AGENTS.md` for agent/import hotspots.
   - `src/channels/AGENTS.md` and `src/plugins/AGENTS.md` for plugin/channel
     laziness.
   - `src/gateway/AGENTS.md` for server lifecycle tests.
   - `test/helpers/AGENTS.md` and `test/helpers/channels/AGENTS.md` for shared
     contract helpers.
   - `src/infra/outbound/AGENTS.md` for outbound/media/action tests.
2. Establish a baseline before changing code:
   - Prefer `pnpm test:perf:groups --full-suite --allow-failures --output <file>`
     for full-suite ranking.
   - For bundled plugin breadth, run the smallest relevant `pnpm
test:extensions:batch <plugin[,plugin...]>` or plugin-inspector command
     before jumping to the full extension sweep.
   - For a scoped hotspot use:
     `/usr/bin/time -l pnpm test <file-or-files> --maxWorkers=1 --reporter=verbose`
   - For import-heavy suspicion add:
     `ASTROCLAW_VITEST_IMPORT_DURATIONS=1 ASTROCLAW_VITEST_PRINT_IMPORT_BREAKDOWN=1`.
3. Separate wall/runner noise from real file cost:
   - Compare Vitest duration, test body timing, import breakdown, wall time, and
     max RSS.
   - Re-run single files when grouped/full-suite numbers look stale or noisy.
   - If a full-suite grouped run reports a lane failure but JSON says tests
     passed, capture that as harness/noise and verify the suspect file directly.
4. Pick the next attack by return and risk:
   - High return: one file/test dominates seconds or RSS and has a clear root.
   - High leverage: one plugin or SDK barrel causes every plugin-inspector or
     extension-batch run to load broad runtime.
   - Lower risk: static descriptors, target parsing, routing, auth bypass,
     setup hints, registry fixtures, or test server lifecycle.
   - Higher risk: real memory/runtime behavior, live providers, protocol
     contracts, or broad production refactors.
5. Fix the root cause, not the symptom:
   - Move static metadata/parsing into narrow helpers or lightweight artifacts
     reused by full runtime and fast paths.
   - Prefer dependency injection, loaded-plugin-only lookup, explicit fixtures,
     and pure helpers over broad mocks.
   - Reuse suite-level servers/clients when a fresh handshake is irrelevant.
   - Keep schedulers/background loops off unless the test proves scheduling.
   - In plugin paths, move static metadata into manifest/lightweight artifacts
     and keep runtime plugin loads behind explicit execution boundaries.
6. Preserve coverage shape:
   - Do not delete a slow integration proof unless the exact production
     composition is extracted into a named helper and tested.
   - Keep one cheap integration smoke when cross-component wiring matters.
   - State explicitly what incidental coverage was removed, if any.
7. Re-benchmark the same command after the change and compute seconds plus
   percent gain.
8. Update the running report when requested or when this thread is tracking one.
   Include before/after commands, artifacts, coverage notes, verification, and
   next attack order.
9. Commit with `scripts/committer "<message>" <paths...>` and push when the
   user asked for commits/pushes. Stage only files touched for this attack.

## Plugin-Suite Workflow

Use this section when perf work involves bundled plugins, plugin-inspector, SDK
barrels, package-boundary tests, or extension suites.

1. Map the suite shape first:
   - source tests: `pnpm test extensions/<id>` or `pnpm test:extensions:batch <id>`
   - package boundaries: `pnpm run test:extensions:package-boundary:canary` and
     `pnpm run test:extensions:package-boundary:compile`
   - all bundled source tests: `pnpm test:extensions`
   - plugin import memory: `pnpm test:extensions:memory -- --json .artifacts/test-perf/extensions-memory.json`
   - plugin-inspector/report work: keep report primitives in `plugin-inspector`;
     keep wrappers thin and collect peak RSS when the command supports it.
2. Start narrow, then widen:
   - one plugin changed: run that plugin's tests and plugin-inspector slice.
   - SDK/public barrel changed: add representative provider, channel, memory,
     and feature plugins.
   - loader/runtime mirror changed: add package-boundary checks and build/package
     proof as needed.
   - unknown shared plugin behavior: run `test:extensions:batch` groups before
     `pnpm test:extensions`.
3. Treat plugin-inspector failures as product signals:
   - JSON must parse.
   - warnings/errors must be classified, not hidden.
   - runtime capture should be quiet and config-tolerant.
   - command output should include wall time, exit code, and peak RSS when
     available.
4. For broad or package-heavy plugin proof, use Crabbox-backed Blacksmith
   Testbox by default on maintainer machines:
   - `pnpm crabbox:run -- --provider blacksmith-testbox --timing-json -- ASTROCLAW_TESTBOX=1 pnpm test:extensions:batch <ids>`
   - add `--keep`/`--id <id-or-slug>` only when several commands must share one
     warmed box; stop it with `pnpm crabbox:stop -- <id-or-slug>`.
5. If plugin performance is package-artifact sensitive, switch to
   `astroclaw-pre-release-plugin-testing` and Package Acceptance rather than
   trusting source-only timing.

## Metric Collection

Collect at least one stable metric before and after. Prefer the same machine and
same command. For Testbox comparisons, use the same `tbx_...` id when possible.

| Metric          | Use for                            | Preferred source                                                            |
| --------------- | ---------------------------------- | --------------------------------------------------------------------------- |
| wall time       | user-visible suite cost            | `/usr/bin/time -l`, test wrapper duration, Testbox run time                 |
| Vitest duration | test body/import cost              | Vitest output per file/shard                                                |
| import duration | broad barrel/runtime loads         | `ASTROCLAW_VITEST_IMPORT_DURATIONS=1`                                        |
| max RSS         | memory pressure and OOM risk       | `/usr/bin/time -l`, `pnpm test:extensions:memory`, wrapper memory summaries |
| CPU/user/sys    | CPU-bound vs wait-bound split      | `/usr/bin/time -l` locally, Testbox job timing when local CPU is noisy      |
| heap snapshots  | real leak vs retained module graph | `astroclaw-test-heap-leaks` workflow                                         |

Local scoped command with CPU/RSS:

```bash
timeout 240 /usr/bin/time -l pnpm test <file> --maxWorkers=1 --reporter=verbose
```

Plugin import memory profile:

From P1R4N351/astroclaw, MIT.

More from P1R4N351/astroclaw