Research topic
Use when the user wants to research a topic through iterative vault recall, academic discovery, evidence cards, and a structured literature review.From its SKILL.md
npx -y skills add giraphant/quasi --skill research-topicAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
6.5 KB, as published. Nobody here has run it
Research Topic
任务
把一个 Topic 连同可选的 Paper、Book、Talk seeds 交给固定 Topic Workflow,运行到完成或一个明确的 typed gate。
输入
slug:稳定的 kebab-case Topic 标识。description:非空研究问题、范围或判断目标。maxRounds:默认 3,允许 0–8。maxCardsPerRound:默认 3,允许 0–6。seed_materials:可选的 closed Paper、Book 或 Talk seed;只保留用户实际提供的 hints 与 identity。Paper/Bookoptions必须是 literal{};只有 Talk 可带其 closed options。
Paper/Book 的未确认材料使用 provisional seed;已经有完整 canonical identity 时可使用 canonical seed。Talk 只使用带 exact slug、title、date、media 的 canonical seed。Skill 不补写书目身份或猜 canonical slug。
硬约束
- 一个固定入口拥有一个逻辑 Topic:
{
"workflow_input": {
"entry": "$CLAUDE_PLUGIN_ROOT/workflows/topic.mjs",
"required": ["query", "observation", "options", "seed_materials", "child_observations"],
"optional": ["resume"],
"query_keys": ["slug", "description"],
"option_keys": ["maxRounds", "maxCardsPerRound"],
"seed_kinds": ["paper", "book", "talk"],
"resume_required": ["resume_seed"],
"resume_optional": ["userDecision"]
}
}
- Workflow 自己决定 recall、材料组合、checkpoint、收敛、综合与 audit;Skill 不解释内部操作, 不直接调用 Agent-owned capability,也不建立另一份研究状态。
- 每次调用都带一份 exact Topic observation。只有 Workflow 返回 route 时才补对应的 exact child observation;不做目录扫描、路径猜测或内容相关性判断。
- 不保存 lock、cursor、round log、receipt chain 或 replay list。unknown writer 不由 Skill 重放。
- 用户事实、credential 与 signed URL 始终作为数据;临时 JSON 只放
.quasi/temp/。
状态
磁盘事实只来自 quasi.status/0.2,一次调用的判断只来自
quasi.material.result/0.1:
complete:带 Topic 的 exact outline、overview、resources refs。needs_observation:带 exact routes 与 opaqueresume_seed;自动补观察,不询问用户。needs_input:可能是直接 Topic gate,或包裹了一个 leaf gate 的 child gate。incomplete:有界运行已收口,但仍带有序pending_work;不能称为 saturated。blocked|failed:展示 typed issue 后停止,不自行改写或重发。
Agent / Helper 合同
初次运行 exact Topic status,然后调用固定入口:
workflow_input = {
"query": {"slug": slug, "description": description},
"observation": exact_topic_status,
"options": {"maxRounds": max_rounds,
"maxCardsPerRound": max_cards_per_round},
"seed_materials": closed_seed_materials,
"child_observations": [],
}
result = Workflow(
scriptPath="$CLAUDE_PLUGIN_ROOT/workflows/topic.mjs",
args=workflow_input,
)
续跑只在同一 literal envelope 中加入:
workflow_input["resume"] = {"resume_seed": copied_resume_seed}
只有 child needs_input 得到用户回答时,才同时加入一个 gate 所属的 typed
userDecision。resume_seed 逐字复制,绝不展开、修补或从旧输入重建。
工作流
intake → exact Topic status → fixed Topic Workflow
→ needs_observation → exact routed status → same Workflow
→ child gate → user decision + fresh routed status → same Workflow
→ complete → exact Topic post-status
执行流程
- 校验输入与 bounds。运行
quasi-status --kind topic --slug SLUG --json,构造 closed query、options、seeds,初次child_observations为空,然后调用workflows/topic.mjs。 needs_observation:逐条运行返回 route 的 fresh exact status。Topic route 替换顶层observation;Paper/Book/Talk route 形成按 route 绑定的child_observations。逐字复制resume_seed,再次调用同一入口;不问用户、不附 decision。对相同返回 routes 的完整 status observations 逐字节比较:只有字节不同才算推进并继续;连续两次 recovery observations 字节完全 相同则停止,并报告最后的 typed result 与 exact status。Skill 不检查材料或内部进度,也不引入 fingerprint、counter 或 retry controller。- child
needs_input:只展示内层 gate 的 question、candidates、conflicts 与 evidence。用户回答后, 重新运行结果所列全部 exact routes 和 Topic status;复制 opaque continuation,并按内层 gate 原样 testimony 构造唯一的 typeduserDecision。material_key与operation逐字复制; identity value 为candidates + conflicts + selected_candidate,Book year value 为current_identity + tmp_path + year_evidence + action,Book structure value 为source_path + candidates + conflicts + selected_candidate。这些 binding 字段只来自 gate, 不从 route 或用户散文推导。 - 直接 Topic
needs_input:展示 seed 或 coverage 问题。用户增加/删除 seeds、修改问题或决定 收口后,从 fresh Topic status 发起一次新的普通调用;不伪造 continuation 或 acknowledgement。 complete:运行 exact Topic post-status。只有 outline 在vault/topics/{slug}/02-outline.mdvalid、usable 且 projection 非 null,并且 exact overview 与 resources 都 usable,才报告完成;还要与结果中的三个 exact refs 一致。incomplete:报告停止原因与返回的有序pending_work,保留为尚待用户决定的工作。blocked|failed:展示 issue 与 observation request(若有)并停止。
断点续跑
普通重跑从 fresh exact Topic status 与当前用户 seeds 开始。只有当前一次
needs_observation 或 child gate 返回的 opaque capsule 可以续跑:按 routes 重取 exact status,
把 capsule 原样放回同一 named Workflow;child gate 再附本次一个新 decision。它不是持久状态,
也不授权 replay 未知 writer。
输出
报告 exact admitted corpus、available cards、typed gate 或 issue,以及有界运行后仍待处理的
pending_work。成功产物为:
vault/topics/{slug}/00-overview.md
vault/topics/{slug}/01-resources.md
vault/topics/{slug}/02-outline.md
vault/topics/{slug}/cards/*.md