362건 · 관련도순
当使用者说"帮我填词"、"按这个字数模板写歌词"、"用多Agent对抗式写词"、"这首副歌帮我想几句"时使用，前提是字数骨架和这首歌写什么都已经有了；只有主题、还没定写什么时先用 lyric-concept-builder。拿一份字数骨架和意象系统，派三个 sonnet 子代理按不同路子并行独立起草（场景派 / 口语派 / 意象派），再派一个红队子代理逐句挑毛病，主 Agent 按句裁决拼稿，最后统一辙口与意象。收敛判据卡死：字数用 lyric-doctor 的脚本对模板数、每段同辙、hook 不含陈词、主歌二对主歌一有推进。不抄现有歌词，不替使用者决定这首歌要说什么。
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Walk every branching path and boundary condition in content, report only unhandled edge cases. Orthogonal to adversarial review - method-driven not attitude-driven. Use when you need exhaustive edge-case analysis of code, specs, or diffs.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Review code changes adversarially using parallel review layers (Blind Hunter, Edge Case Hunter, Acceptance Auditor) with structured triage into actionable categories. Use when the user says "run code review" or "review this code"
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Run multiple Reforge-guided section-refactor lanes in parallel, each in its own worktree/branch off origin/main, with a score-blind adversarial review panel gating every commit, one draft PR per section. Use when the user wants to reduce Reforge surface/internal score across several sections at once via architecturally-correct deletions (dead surface, entity-leak removal, cross-section consumers moved onto the public contract) — not by relocation, parameter bags, or accessibility-dodging. Wraps the per-lane process in .codex/skills/humans-refactor. Triggers: 'refactor swarm', 'parallel section refactors', 'run reforge refactor on Users/Tickets/...', 'burn down surface score across sections'. Has an intensity dial (light|standard|deep) and a workflow|solo execution mode to trade token burn against autonomy per unit of output.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Adversarial triage of open review findings on a PR. Every finding (Codex, Claude bot, Gemini) is a hostile hypothesis to REFUTE — fix only what is proven real, reachable, in the PR's scope, and above the round's severity bar. The bar rises every review round. Everything else is declined or filed as an issue. Use for 'fix the PR findings', '/fix', '/fix 123', or before any fix commit in a review loop.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Orchestrate autonomous AI development pipelines through your Kanban board (Asana, GitHub Projects, Linear). Manages multi-worker Claude Code dispatch, deterministic quality gates, adversarial review, per-task cost tracking, and crash-proof pipeline execution.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Adversarial thesis testing — act as a short seller to find the weakest points in a thesis. Use when user says "stress test", "devil's advocate", "challenge this thesis", or "break [TICKER]".
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Adversarial verification of one SDD Composy task in verify_required: reproduce every claim with fresh commands, build the truth table, refute stale evidence, and produce the verdict that gates complete. Use when review is done and the task claims to be finished — 'verificar a TASK-003', 'a task está pronta mesmo?', 'is this task really complete'. Do NOT use for QA, code review, or a general 'verify my work' outside an SDD task.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Use when about to claim work is complete, fixed, or passing, and when a finished feature needs adversarial verification before integration
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Adversarially verify an implementation against its objective, acceptance criteria, definition of done, and prohibitions using fresh evidence. Use for completion gates, release readiness, or independent verification after implementation and review.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Deliver a small, bounded code change through inspection, a mini-plan, implementation, testing, review, and adversarial verification. Use for bug fixes, configuration changes, and features expected to touch no more than five files.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
OpenAI Codex CLI wrapper — three modes. Code review: independent diff review via codex review with pass/fail gate. Challenge: adversarial mode that tries to break your code. Consult: ask codex anything with session continuity for follow-ups. The "200 IQ autistic developer" second opinion. Use when asked to "codex review", "codex challenge", "ask codex", "second opinion", or "consult codex". (gstack) Voice triggers (speech-to-text aliases): "code x", "code ex", "get another opinion".
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Score an unbuilt business idea on desirability, viability, feasibility, distribution wedge, why-now, founder-market-fit, market size, alternatives, defensibility, capital intensity, and regulatory/ethical risk — and return a GO / ITERATE / KILL verdict with kill criteria and a next kill test. Load when the user asks to evaluate a business idea, score a startup idea, screen an idea, decide whether to pursue this venture, do an idea review, or says "is this a good business idea", "should I build this", "evaluate this startup", "screen this idea", "go/no-go on this idea", "kill or pursue". Sub-skill of `venture-exploration`. Calls `fermi` for sizing, `assumption-mapping` for hidden beliefs, optional `pre-mortem` / `adversarial-hat` for high-stakes ideas. Does NOT evaluate built products — for that use `reality-check`.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Orchestrate one or more thinking frameworks to work through any problem, decision, document, or idea rigorously. Diagnoses which frameworks fit — inversion, pre-mortem, assumption-mapping, socratic, adversarial-hat — then guides the user through them in the right sequence. Load when the user asks for deep thinking, says "help me think through this properly", "apply your best thinking frameworks", "I need to think carefully before deciding", or "what thinking tools should I use here". Also the entry point for any complex problem where the right framework is unclear. Covers product decisions, engineering tradeoffs, personal decisions, strategy, creative challenges — any domain.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Classify every destructive/gated action an agent body can attempt (git, filesystem, network, shell, github) into block/approve/allow tiers, and audit whether the pre-tool and post-tool governance backing that classification is real: blocked actions proven zero-side-effect by a negative fixture, every denial backed by a receipt and a transcript event, every block-tier denial paired with a safe alternative, and no unmanaged or same-UID body ever marked "contained." Use when building or reviewing Agent Harbor's C5 governance gate (destructive git blocker, approval request/result, denial receipts), gating a new tool surface before it can run destructive commands, or auditing whether an existing "destructive-action blocked" claim actually has evidence behind it. NOT for proving an existing sandbox boundary actually contains an adversary once an action is isolated (use sandboxed-adversarial-test-harness), deciding where in a DAG to place a human review node or designing the approval UX itself (use human-gate-designer), or designing the general work-receipt schema across an entire agent task (use agent-work-receipt-designer).
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Treat tokens as a coding swarm's cost-of-goods-sold and legibility engine while accounting for finite subscription allowance when per-call price is unknown. Covers provider windows, remaining-usage evidence, burn forecasts, preemptive checkpoint/model switching, per-agent budgets, compaction, shared digests, context degradation, and spend metering. Activate on: "token budget", "context budget", "compaction strategy", "briefing as compression", "context rot", "summarization collapse", "COGS for agents", "subscription usage remaining", "five-hour or weekly limit", "burn forecast", "model switch before limit", or "/context-economics-for-agent-swarms". NOT for: memory architecture (use always-on-agent-architecture), single-prompt wording (use prompt-engineer), mechanism-design proofs (use nisan-et-al-2007-algorithmic-game-theory), physical containment (use sandboxed-adversarial-test-harness), or production rebodiment and fencing (use agent-resurrection-and-body-continuity).
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Audit a compliance-ladder design (C0-C6) and transcript-fidelity ladder (T0-T5) plus their adapter conformance fixtures — Codex, Claude Code, Cloudflare, Ollama/LM Studio, custom stdio/HTTP agents — for the two failure modes that make a C-badge or T-fidelity label a lie: the ladder disagreeing with itself across doc/schema/ui/probe surfaces, and a level being reachable by self-report because no daemon-witnessed negative probe actually tries to forge it. Use when designing or freezing a compliance ladder, wiring an adapter conformance probe suite, deciding whether a numeric C-badge or T-label is safe to ship, or investigating a self-attestation or observed-to-controlled bypass risk. NOT for auditing a cryptographic protocol or whitepaper for adversarial soundness (`redteam-review`), writing TLA+/formal invariants for a daemon monitor (`runtime-verification-for-agents`), or agent identity, successor linkage, and reputation across restarts (`agent-identity-continuity-reputation`).
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Extract exactly what the person typed in a Claude Code session (script, no agent context spent), have a fresh agent write a faithful structured summary under it, and optionally have an adversarial agent judge whether the working agent is still on track. Use at a checkpoint, before a handoff, after a compaction, when a long session may have drifted from what the owner asked, or when told to 'summarize what I said' or 'check we're on track'.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Pre-commitment attack on a plan by an agent that did not write it. Use when presenting a feature-deep-dive plan, assigning vision-to-fleet briefs, moving a register to build mode, ratifying a scope list, or coding a new contract or table. NOT for built work or DONE reports (use subagent-dispatch).
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Real tests: a fresh agent walks the live product against sealed checks proven to catch a planted break. Use when told to use safety-net, to check whether an area really works now, or around a change too wide for one agent to see every break. NOT for one automated test (use forcing-function-tests).
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
This skill should be used when the user wants to "test a claim against data", "validate an empirical hypothesis", "refine a model/theory with evidence", "characterize a pattern and verify it", "演繹と帰納で検証", "仮説をデータで検証して更新", "実例と反例で確かめる", "弁証法ループ", "主張を現物で裏取り", "傾向分析を検証して精緻化", or mentions iteratively updating a hypothesis by deriving predictions and testing them against a real corpus. NOTE: Use this for VALIDATING/REFINING an empirical claim or model against real data through a derive→test→update loop. Use strong-inference for debugging an UNKNOWN cause, and devils-advocate for adversarially stress-testing a DESIGN proposal.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Turn a one-line objective into a step-by-step construction plan for multi-session, multi-agent engineering projects. Each step has a self-contained context brief so a fresh agent can execute it cold. Includes adversarial review gate, dependency graph, parallel step detection, anti-pattern catalog, and plan mutation protocol. TRIGGER when: user requests a plan, blueprint, or roadmap for a complex multi-PR task, or describes work that needs multiple sessions. DO NOT TRIGGER when: task is completable in a single PR or fewer than 3 tool calls, or user says "just do it".
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Orchestrate autonomous AI development pipelines through your Kanban board (Asana, GitHub Projects, Linear). Manages multi-worker Claude Code dispatch, deterministic quality gates, adversarial review, per-task cost tracking, and crash-proof pipeline execution.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.
Three-layer verification pipeline for AI output. Extracts verifiable claims, finds supporting or contradicting sources via web search, runs adversarial review for hallucination patterns, and produces a structured verification report with source links for human review.
한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.