테스트·보안 스킬

Family-G proof & derivation integrity forensics: does a THIRD PARTY's written proof/derivation actually establish its theorem, or does it skip an obligation, assume its own conclusion, take an invalid step, drift a symbol's meaning, or smuggle an unstated assumption? Decides from the WRITTEN proof/derivation — verdict-bearing at L1 (the LaTeX source; PDF-extracted math is unreliable, so an L0 PDF-only run surfaces info only) — never asserts 'fabricated', only that the step shown does not hold. A fresh cross-model reviewer reads the theorem/proof + an extraction-only obligation scaffold and proposes per-obligation findings, each span-anchored to the evidence ledger (claims.json); reviewer≠adjudicator. Emits proof-derivation-forensics.findings.json; NEVER computes the verdict. dimension=proof, can be critical. Triggers: "proof forensics", "check this proof", "derivation integrity", "audit the math", "证明审计", "推导有没有漏洞".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Flagship intra-paper self-consistency forensics: does the paper contradict ITSELF across abstract/intro/tables/body/appendix, and does the method DESCRIBED match the method EVALUATED? Needs no external ground truth — works PDF-only (L0). Runs a deterministic arithmetic pass + a fresh cross-model semantic pass, every finding span-anchored to the evidence ledger (claims.json), reviewer≠adjudicator. Emits consistency-audit.findings.json; NEVER computes the verdict. Triggers: "consistency audit", "check the paper against itself", "self-consistency", "内部自洽".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audit whether a paper's baseline comparisons are COMPLETE, FAIR, and SIGNIFICANT: a required recent SOTA baseline is missing while 'best/SOTA' is claimed (HP-MISSING-BASELINE); a baseline is undertuned / given less compute-tuning-data, run at a mismatched config, or the equal-budget ablation-as-baseline is absent (HP-WEAK-BASELINE); 'outperforms' is asserted over overlapping error bars or with no variance/seeds (HP-SIG-OVERLAP); and a cross-row 'improves over baseline by X%' is arithmetically wrong (HP-DELTA-ERROR, cross-row form only). A versioned per-domain baseline profile + a live leaderboard/recency search are assembled by the EXECUTOR as structured facts; a fresh cross-model reviewer (gpt-5.6-sol xhigh, read-only, fresh thread per dimension) PROPOSES findings, each span-anchored to a ledger claim_id; tools/adjudicate_findings.py DECIDES the verdict. Works at L0 (stated comparisons) and deepens at L2 (configs/result files). A completeness question it cannot settle internally becomes needs_external_check, never a guessed missing baseline. Emits baseline-comparison-audit.findings.json; computes NO verdict. Detect-only. Triggers: "baseline audit", "missing baselines", "is the comparison fair", "weak baseline", "baseline 误报", "SOTA earned?".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Synthesize the single strongest EVIDENCE-BOUND reviewer case to reject a paper, built ONLY from the evidence ledger (claims.json) + the other auditors' confirmed findings — never free-floating LLM critique. Two fresh cross-model codex threads: an attack writes the ~200-word rejection paragraph (every accusation tagged to an existing claim_id/finding_id), a defense decomposes it and rules each point against the anchored evidence. MEMO-ONLY: emits adversarial-case-builder.memo.md (fed to the adjudicator via --memo) and carries NO verdict weight — tools/adjudicate_findings.py lists it in ZERO_WEIGHT_SKILLS and caps it at info. Honest-null allowed (the paper may survive). Run LAST. Detect-only. Adapted from ARIS kill-argument. Triggers: "adversarial case", "strongest objection", "rejection memo", "kill argument", "最强拒稿点".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Build the deterministic evidence ledger (artifact_manifest.json + claims.json) that every other Anti-Autoresearch auditor reads. One pass inventories artifacts, derives the observability level (L0 PDF-only / L1 +LaTeX / L2 +repo+results) by fixed rule, and extracts span-anchored, hashed, checkable claims (numbers, comparisons, scope, method, baselines, citations, captions, table cells) into claims.json. An OPTIONAL additive cross-model pass ADDS span-anchored semantic claims — method, theorem statements with their assumptions, definitions, proof/derivation steps and equations, scope, baselines, conclusions, the motivation span, and reproducibility-artifact references (the proof, derivation, and structure anchors the family B/D/G auditors need) — it never invents a number, emits a finding, or computes a verdict. Run FIRST, before any audit skill. Triggers: "build the ledger", "extract claims", "prep for integrity audit", "evidence ledger", "建证据账本".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Transparent, itemized impressions of AI-generated WRITING STYLE — the repo's ONLY non-integrity track. Two passes: a deterministic defensive-hedge density screen (tools/check_ai_style.py, AIS-DEFENSIVE-HEDGE) plus a fresh cross-model GROSS-cases-only semantic pass over the 13 AIS-* style tells (broken narrative arc, LLM phrase tics, jargon-stuffing, invented codenames, clause/formula walls, gratuitous pseudocode, bullet overuse, bold-module spam, restatement loops, focus drift, single-style figures, appendix dumping). Every finding is named, LOCATED, span-anchored to the evidence ledger (claims.json), carries not_integrity_finding:true + false_positive_risk:high + an fp_case, and gets ZERO verdict weight: the adjudicator forces it to info, excludes it from overall_verdict, and renders it in a SEPARATE report section. NOT an AI-text classifier — no scores, no "this is AI-written", no authorship probability; a paper can be CLEAN_GIVEN_EVIDENCE and still list many. Emits ai-style-impressions.findings.json; computes NO verdict; if a tell is actually substantive it routes to the integrity auditor. Triggers: "AI style", "vibe check", "writing fingerprint", "AI 文风", "vibe paper".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audit experiment integrity against the evidence ledger. At L2 (repo + result files present) a fresh cross-model reviewer reads the eval code line-by-line for fake/derived ground truth, score self-normalization, phantom results (a paper number with no backing file/key), dead/uncalled metric code, verified-scope inflation, method-described ≠ method-evaluated drift, synthesized-looking results, placeholder/fake data still wired into a released result, code-output ≠ reported-number mismatch, and missing reproducibility artifacts (an empirical/agent/LLM paper shipping neither code nor the prompts/configs its results need) — every finding span-anchored to a ledger claim_id. At L0/L1 (PDF / source only) the same patterns are surfaced as info-level 'could-not-verify' signals where the ledger gives an anchor (observability_level_required:2) — NEVER a fraud verdict from a PDF. The reviewer PROPOSES findings; tools/adjudicate_findings.py computes the verdict. Detect-only. Triggers: "experiment forensics", "audit the results", "check the eval code", "实验诚实度".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Black-box QA audit of slop-guard across MCP, CLI, fit, docs, agent workflows, and writing-effectiveness. Files GitHub issues for real problems found.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Calibrated grilling session for stress-testing a plan, design, idea, or decision. First assesses the user's topic knowledge, confidence, and desired pressure level, then asks one question at a time with recommended answers. Use when user says "grill me", "stress-test this", "challenge my plan", "interview me", or wants a plan probed without being overwhelmed.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audits font loading, type scales, measure, spacing, OpenType, and rendered punctuation with 78 scoped rules. Use when asked to "audit typography", "fix the fonts", or "review my type system". For a new visual direction use ui-design Direction; for general UI defects use ui-design Audit.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Reviews a local diff, branch diff, or PR with file:line findings in confirmed and plausible tiers, gates any test the diff adds or changes, and in apply mode lands the fixes and diff-scoped simplifications. Use when asked to "review my changes", "code review", "tidy this", "simplify my diff", "deslop this", "structural review", "security audit", or "audit these tests in my PR". For the PR itself use pr-creator; for CI and review threads use pr-babysitter; for UI defects use ui-design; for repo architecture use codebase-architecture; for pruning the whole suite to a target use test-audit.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Prunes low-value tests across a whole suite to a measured target while holding coverage. Use when asked to "remove useless tests", "prune the test suite", "we have too many tests", "agents keep writing pointless tests", or "cut 20% of tests without losing coverage".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Creates and reviews executable implementation plans grounded in repository evidence, with vertical slices, explicit decisions, and verification criteria. Use when asked to "plan this feature", "stress-test this plan", "grill me", or "split this into tickets". For architecture use codebase-architecture; for code review use tidy.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audits libraries, CLIs, and SDKs using 38 rules for public contracts, package exports, piped output, errors, and configuration. Use when asked to "audit my CLI", "review my SDK", or diagnose package type resolution.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Speeds up CI and the merge path: finds where changes wait between push and deploy, then fixes the critical path, check placement, merge queue or post-merge checks with auto-revert, flaky tests, and deploy rate. Use when "CI is slow", "speed up CI", "do we need a merge queue", "main keeps going red", "flaky tests block merges", agents open PRs faster than they merge, or building a self-healing merge loop.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audits agentic products for tool parity, authority, approval payloads, recovery, and trust using 24 rules and a ship verdict. Use when asked for an "AX audit", to review an agent approval flow, or whether an agent can operate the product.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Audits and edits agent instruction files, verifies repository commands, and migrates repositories to AGENTS.md as the single shared source. Use when asked to "improve my AGENTS.md", "migrate CLAUDE.md to AGENTS.md", or make instructions work across agents.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Creates and improves portable Agent Skills with a validator, routing scenarios, and evidence-based keep, cut, merge, or retire decisions. Use when asked to "write a skill", "update all skills", "audit my SKILL.md", "remove redundant instructions", or fix skill triggering.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Keeps specification documents and code in agreement. Audit mode finds every divergence when an implementation is claimed complete; sync mode detects spec drift in staged changes and updates the spec after user approval, so each commit is a reconciled snapshot of spec, tests, and code. Use when an implementation is claimed complete against a spec, before committing changes that may drift from the spec, or after approving decisions that must flow back into it.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Generates a behavioral scenario acceptance matrix for comprehensive test coverage planning. Maps user scenarios to acceptance criteria with stable IDs. Use when planning test coverage for a complex feature, before a release or milestone review, or when someone asks what is tested.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Runs the Implement→Review→Test→Fix cycle until the work is actually clean, with evidence. Use after any non-trivial implementation and before claiming it is done.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Strips AI-generation "smell" from prose before it ships (READMEs, docs, release notes, PR descriptions, posts, emails). Counts measurable tells first (keyword saturation, tricolons and anaphora, the "not X, it's Y" reflex, fragment-for-emphasis, bolded thesis lines, listicle-ization, signpost phrases, filler vocabulary, em-dashes, engineered-quotable closers), then applies a "doing work or reflex?" test to each instance so genuine voice survives. Use when asked to de-slop text or make it sound human, when text reads as machine-written, and as the final pass on prose the agent drafted.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Scans staged changes for secrets, debug statements, TODOs without references, and other common commit mistakes. Use before creating any commit.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

Generates edge case, failure mode, and spec-driven test cases. Covers boundary values, nil inputs, concurrency, resource exhaustion, malformed data, and requirement-linked traceability tests. Use after happy-path tests exist and before claiming coverage is complete, when requirements lack tests, or before a security review.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.