blackstone2333
UserGoldilocks：精简的 Superpowers 替代方案｜v0.4.1 Direct 对照 114/114 通过，处理 token −11.5%、耗时 −10.9%、官方成本 −6.3%。Lean, token-efficient AI agent workflow for Codex & Claude Code.
Categories
Indexed Skills (15)
goldilocks
Adaptive workflow and quality-control router for any executable work, including software, research, analysis, documents, presentations, spreadsheets, and other structured deliverables. Invoke implicitly or explicitly to choose Direct work, alignment, diagnosis, planning, delegation, continuity, or verification; clear tasks immediately take the Direct exit without loading another workflow reference.
subagent-driven-development
Use when executing a plan with multiple coherent implementation units and available subagents can reduce elapsed time or context load without weakening integration quality.
artifact-production
Use when a structured artifact such as a presentation, report, long document, workbook, or edited video has separable units that benefit from shared design, parallel production, localized rework, and one controlled integration path.
brainstorming
Use when the user explicitly asks to brainstorm, design, explore product or architecture options, clarify a materially ambiguous idea, or define an end state before implementation.
dispatching-parallel-agents
Use when two or more independent workstreams can run concurrently without shared mutable state or sequential dependencies and parallelism has clear net benefit.
executing-plans
Use when a written implementation plan already exists and the user wants it reviewed and executed with appropriate checkpoints, tests, and completion evidence.
finishing-a-development-branch
Use when implementation is verified and the development branch or worktree must be handed off, pushed, merged, proposed as a pull request, retained, or safely cleaned up.
receiving-code-review
Use when code review feedback must be evaluated, clarified, accepted, rejected, or implemented, especially when a suggestion is vague, risky, or technically questionable.
requesting-code-review
Use when the user asks for code review, a significant or high-risk change needs an independent pass, or work is approaching integration and review has positive quality value.
systematic-debugging
Use whenever the user asks to debug, diagnose, investigate, troubleshoot, or fix a bug, failing or flaky test, build error, performance regression, integration mismatch, or unexpected technical behavior; establish root cause before proposing changes.
test-driven-development
Use when implementing a bug fix or non-trivial behavior whose acceptance can be expressed as a meaningful failing test before the production change.
using-git-worktrees
Use when feature work needs isolation, concurrent branches must not share state, a dirty baseline must be protected, or the user explicitly requests a git worktree.
verification-before-completion
Use before consequential or multi-claim completion, safety, integration, release, or deployment handoff when evidence needs deliberate coverage. Do not invoke for a clear Direct edit solely because the request says verify, complete, or fixed; run its one obvious targeted check inline.
writing-plans
Use when requirements are sufficiently settled and a multi-step implementation plan, durable handoff, integration sequence, or risk-controlled execution design is needed before coding.
writing-skills
Use when creating, editing, testing, reviewing, or deploying an agent Skill, especially when its triggers, token cost, behavioral effect, or progressive-disclosure structure must be validated.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.