AgentStack
Browse Sign in
Browse Why AgentStack Sell Docs
Sign in
SKILL verified MIT Self-run

Stimuli Exp Design Review

skill-cuteweather-agent-stimuli-exp-design-review · by cuteweather

在 run_exp_design 与 run_codegen 之间执行阻断式 exp_design 审核,输出机器可读 gate 与人类可读报告,确保设计规格可执行、可追溯、可审计。

No reviews yet
0 installs
40 views
0.0% view→install

Install

$ agentstack add skill-cuteweather-agent-stimuli-exp-design-review

✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.

Security review

✓ Passed

No issues found. Passed automated security review. · v0.1.0 How review works →

  • Prompt-injection patterns
  • Secret / credential exfiltration
  • Dangerous shell & filesystem operations
  • Untrusted network calls
  • Known-malicious package signatures

What it can access

  • Network access No
  • Filesystem access No
  • Shell / process execution No
  • Environment & secrets No
  • Dynamic code execution No

From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.

View the full security report →

Verified badge

Passed review? Show it. Paste this badge into your README, it links to the public security report.

AgentStack Verified badge Links to your public security report.
[![AgentStack Verified](https://agentstack.voostack.com/badges/verified.svg)](https://agentstack.voostack.com/security/report/skill-cuteweather-agent-stimuli-exp-design-review)

Reliability & compatibility

Security review passed
0 installs to date
no reviews yet
4mo ago

Declared compatibility

Claude CodeClaude Desktop

Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.

Preview Execution monitoring

We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.

How agent discovery & health will work →
Are you the author of Stimuli Exp Design Review? Claim this listing to set pricing, connect Stripe payouts, and keep 70% of every sale.
Sign up to claim

About

Stimuli Exp Design Review

Purpose

你只做一件事:审核 /exp_design.md 是否达到可进入 code generation 的标准。

Inputs (Priority)

1) /exp_design.md 2) /paper/ 3) /data/ 4) /repro_check.md(若存在)

Required Outputs

  • /exp_design_review.md
  • /script/exp_design_review_gate.json

Review Method (Blocking)

必须沿用既有 exp_design 审核思路,至少覆盖以下检查项:

  1. Evidence Index 完整性:关键事实都能追溯到明确证据定位。
  2. Parameter Registry 完整性:关键实现参数可落地、无明显缺项、无自相矛盾。
  3. Fact Ledger 一致性:事实陈述与 paper/data/code 证据一致。
  4. Phase/condition 语义一致性:phase 职责、条件分支、trial 流程可执行且无错位。
  5. Data mapping 完整性:数据字段到实现参数的映射可闭环。
  6. Conflict/Missing 处置:冲突有裁决,缺失有明确阻断说明。
  7. 外部素材策略一致性:若实验依赖外部视觉素材,策略、追溯要求与 trial 映射完整。
  8. Grounding 产物一致性/paper/grounding/grounding_index.json 必须存在于磁盘上(无论有无参考图)。
  • 若 paper 中存在可用于 grounding 的参考图,则 expdesign 阶段应已调用精确技能名 stimuli-grounding-segmentation;若产物缺失或全部回退成 textonly,应优先怀疑 stimuli-grounding-segmentation 没被正确调用或没有被继续重试。
  • 若文件中 status = "no_reference_figures"families 为空数组:确认 paper/ 中确实无图片文件,exp_design.md 中所有 family 的 effective_reference_basis 均为 text_only
  • 若 paper/ 中存在图片文件,且 grounding_index.json 中所有 family 的 effective_reference_basis 都是 text_only:必须进一步检查这些图片是否真的全是结果图/统计图。只要任一图片中可见 stimulus exemplar、procedure panel、layout scene、response dial、search display 或 memory sample,就必须判为 issue,因为这说明 exp_design 错误跳过了 figure grounding。
  • “刺激可由数学参数生成”不能作为通过理由。若 paper 中仍有可见刺激图或流程面板,review 必须要求至少保留 1 条 figure_then_text 的 grounding family;并且尽可能产出crop_and_text 的 grounding 产物,而不是直接回退到 text_only
  • 有参考图就必须产图:若任何 family 的 reference_priority=figure_then_text,则最终产物中必须至少存在 1 张图片产物(mask / overlay / example)。若全部为空,判定为 critical issue,不得以任何理由通过(包括“刺激可由文本参数指定”“无需外部资产”等)。
  • 空产物判定流程(当所有有参考图 family 的产物均为空时):
  1. 检查 retry_attempts 是否非空 → 若空,critical issue(未使用 --retry-sweep
  2. 检查 retry_attempts 中是否包含 phase=fallback_prompt → 若无,issue(未写入 fallback_prompts
  3. 检查是否存在 reference_role=layout_referencematch_status=matched 的兜底记录 → 若无,issue(未做 scene-level 兜底)
  4. 只有当 1-3 全部通过(即已充分重试、已写 fallback prompts、已尝试 scene-level)且仍然无产物时,才可记为 non-critical issue 并允许 codegen 基于文本参数继续
  • 严禁将"有参考图但全部分割失败"标记为 no_reference_figures。若 grounding_index.json 中存在 family 条目(families 非空),则 grounding_status 必须为 has_families,不可为 no_reference_figures
  • 若文件中包含 family 条目且全部 match_status = "no_match"(即有参考图但全部分割失败):判定为 issue(非自动 pass),要求在 issues 中记录失败原因,并验证 exp_design 阶段已尝试至少一次重试或 prompt 调整。
  • 必须检查 retry_attempts 字段:若 family 为 no_matchretry_attempts 为空数组或不存在,说明未使用 --retry-sweep,判定为 critical issue,要求重新运行 stimuli-grounding-segmentation 并带上 --retry-sweep
  • 必须检查 fallback_prompts 覆盖:若 retry_attempts 中没有任何 phase=fallback_prompt 的条目,说明未在 grounding_requests.json 中写入 fallback_prompts,判定为 issue,要求补充 fallback_prompts(至少包含 1 条 scene-level prompt)并重跑。
  • 必须检查 scene-level 兜底:若所有 object-level family 均 no_match,但没有任何 reference_role=layout_referencematch_status=matched 的 family 记录,判定为 issue,要求增加 scene-level grounding 并重跑。
  • 若文件中包含 family 条目但全部 effective_reference_basis = text_only,同时 paper 图中又存在 stimulus/procedure/layout 画面:判定为 issue,要求重新运行 exp_design / grounding,并补出grounding图。
  • 若文件中包含实际 families 条目且任何 effective_reference_basiscrop_and_text:验证 paper/grounding/examples/paper/grounding/overlays/paper/grounding/masks/ 中有实际图片文件。若声称 crop_and_text 但无对应图片,判定为 critical issue
  • grounding_index.json 不存在,判定为 critical issue(说明 exp_design 阶段未执行 grounding 扫描)。

Gate Contract (Mandatory)

script/exp_design_review_gate.json 必须包含以下最小结构:

{
  "passed": true,
  "summary": "string",
  "checks": {
    "executed": true,
    "evidence_index": {"passed": true, "issues": []},
    "parameter_registry": {"passed": true, "issues": []},
    "fact_ledger": {"passed": true, "issues": []},
    "phase_condition_semantics": {"passed": true, "issues": []},
    "data_mapping": {"passed": true, "issues": []},
    "conflict_missing": {"passed": true, "issues": []},
    "asset_strategy": {"required": false, "passed": true, "issues": []},
    "grounding_artifacts": {"required": true, "passed": true, "issues": [], "grounding_status": "has_families | no_reference_figures", "retry_sweep_verified": true}
  },
  "critical_issues": [],
  "next_actions": []
}

Hard Rules

  • critical_issues 只要非空,passed 必须为 false
  • 若任一关键检查未执行,passed 必须为 false
  • exp_design_review.md 与 gate 结论必须一致,不得一个通过一个不通过。
  • 不得用“看起来差不多”替代可核验证据。

Return Contract

按调用方要求返回结构化 JSON。

  • exp_design_review_gate.json.passed = true,可返回 succeeded
  • passed = false(发现阻断问题),返回 failed_retryable,并在 error/summary 明确 blocker。

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

Install and usage instructions live in the source repository linked above.

Reviews

No reviews yet, be the first.

Versions

  • v0.1.0 Imported from the upstream source.