AgentSkillsCN

vibeship-optimizer

适用于任何代码库的安全、可回滚优化工作流：在优化前/后分别截取快照，对比性能、大小与健康探针数据，并长期保存多日的验证日志。适用于 OpenClaw 中，当用户希望在不破坏现有代码的前提下优化项目、实现“每次变更单独提交并可回滚”的版本控制，或希望建立一份实时更新的 VIBESHIP_OPTIMIZER.md 验证文档时使用。

SKILL.md
--- frontmatter
name: vibeship-optimizer
description: Safe, rollback-friendly optimization workflow for any codebase: capture before/after snapshots, compare performance/size/health probes, and maintain a multi-day verification log. Use in OpenClaw when a user wants to optimize a project without breaking it, wants commit-per-change rollbacks, or wants a living VIBESHIP_OPTIMIZER.md validation document.

vibeship-optimizer

Use the vibeship-optimizer CLI from this repo to run a Carmack-style optimization loop.

First run (recommended)

  • •python -m vibeship_optimizer init --onboard --no-prompt

This creates:

  • •VIBESHIP_OPTIMIZER.md (logbook)
  • •.vibeship-optimizer/ (state dir: config, snapshots, changes, reports)

Standard loop (one optimization)

  1. •Start a tracked change (appends a section into VIBESHIP_OPTIMIZER.md):
  • •python -m vibeship_optimizer change start --title "<change title>" --risk "<risk>" --rollback "git revert <sha>"
  1. •Take baseline snapshot (and attach it to the change record):
  • •python -m vibeship_optimizer snapshot --label before --change-id <chg-...> --as before
  1. •

    Make exactly one optimization + commit it.

  2. •

    Take after snapshot (and attach it to the change record; also records current commit sha):

  • •python -m vibeship_optimizer snapshot --label after --change-id <chg-...> --as after
  1. •Compare and write a report:
  • •python -m vibeship_optimizer compare --before <before.json> --after <after.json> --out reports/vibeship_optimizer_compare.md

Multi-day verification

  1. •Start monitoring (baseline defaults to latest snapshot if omitted):
  • •python -m vibeship_optimizer monitor start --change-id <chg-...> --days 5
  1. •Run once per UTC day:
  • •python -m vibeship_optimizer monitor tick

This appends "Verification update" blocks to VIBESHIP_OPTIMIZER.md and stores reports under .vibeship-optimizer/reports/.

Preflight + hallucination protections

  1. •Run diligence checks:
  • •python -m vibeship_optimizer preflight --out reports/vibeship_optimizer_preflight.md
  1. •Evidence-based review before claiming "this helped":
  • •python -m vibeship_optimizer review bundle --change-id <chg-...> --out reports/vibeship_optimizer_review_bundle.md
  1. •Record review attestation (required by default via config):
  • •python -m vibeship_optimizer review attest --change-id <chg-...> --tool codex --reasoning-mode xhigh --model "<model>" --reviewer "<name>"

If you want to relax attestation enforcement, set:

  • •.vibeship-optimizer/config.yml (or .json) -> review.require_attestation: false

Read-only analyzers (safe)

  • •python -m vibeship_optimizer analyze --out reports/vibeship_optimizer_analyze.md

This produces bloat/size hints and naive "maybe unused dependency" hints (heuristic; do not auto-remove without verification).

Automation (OpenClaw cron)

For scheduled daily verification ticks, see:

  • •references/openclaw_cron_setup.md

For cron/automation, prefer:

  • •python -m vibeship_optimizer autopilot tick --change-id <chg-...> --force --format json --ok-on-pending