harness-scaffold

Regenerate the structural skeleton of a harness — agent role stubs, KB layout, and per-stack rule files — for a repository's stack, in seconds. Use when a harness structure must be (re)created from artifacts on disk, after the roles and contract are already understood and never before. Produces structure only: every stub is intentionally empty, and the scaffold grants no authority.

跳到安装

仓库: luanmorenommaciel/uc-transact-co
最近来源活动: 2026年8月15日 00:51
检测到的 SKILL.md 语言: 英语
星标: 49
分支: 48

安装方式

默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

直接命令不会经过审查 Prompt；运行前请先检查来源。

npx skills add https://github.com/luanmorenommaciel/uc-transact-co --skill harness-scaffold

命令会保持在同一行。复制前请横向滚动并检查完整内容。

想先保存到本地？可下载 SkillsMP 当前能够提供的文件。

检查来源文件

决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。

文件资源管理器

2 个文件

SKILL.md

来源说明 · 只读预览

返回搜索返回安装

spec-before-build

Turn one reasonable-looking plan item into a single atomic Task-Spec that states its goal, the behaviour a named human signs, and the machine-runnable proof that answers "am I done?" without asking. Use when a backlog item, plan row, or ticket permits more than one defensible implementation; before handing work to an agent; when "done" is still a matter of opinion; or when a graph of small units must be derived from a large plan. Refuses to invent a definition for any term whose meaning is owned by someone else.

2026-08-14

interview-the-system

Turn an ambiguous question about an unfamiliar or brownfield system into a bounded, read-only, evidence-backed investigation with a confirmed investigation contract, deliberate context manifest, claim ledger, ontology, owned open questions, and reviewable technical brief. Use when asked to investigate a repository, database, service, business metric, or existing system before building; grill a vague request; distinguish physical data from business meaning; or produce grounded upstream context for later specification and planning.

2026-08-13

查看这个仓库的全部 Skills

Harness Scaffold

One run regenerates the harness structure the team already understands: agents/ role stubs, kb/<stack>/ knowledge-base placeholders, and rules/<stack>.md rule stubs, plus a manifest. It is the automation of the hand-written work from sessions/d2/03-agent-pair.md — faster, never smarter.

Day 2 of Semana Engenharia Agêntica delivers this skill. It follows Day 1's interview-the-system, which supplies the evidence and the owned questions the harness is built around, and it precedes Day 3's spec-before-build, which writes the units that execute inside the authority this structure holds.

Boundaries

Writes ONLY under tmp/harness-scaffold/. Never touches AGENTS.md, CLAUDE.md, dbt/, src/, or storage/.

Grants no authority. The harness contract does that; this skill only lays out where the pieces of a harness live.

Every generated file is a stub, empty by design. What fills a stub — grounded KB content, tuned doctrine — is human work measured in hours, not a generator's output.

Requires no chat memory: everything it needs is on disk.

Method

Read the stack from the request (default for this repo: dbt,duckdb,fastapi,mcp).

Run the generator:

uv run python skills/harness-scaffold/scripts/scaffold_harness.py \
  --stack dbt,duckdb,fastapi,mcp --dest tmp/harness-scaffold

Validate the structure:

uv run python skills/harness-scaffold/scripts/scaffold_harness.py \
  --stack dbt,duckdb,fastapi,mcp --dest tmp/harness-scaffold --check

Report: the generated tree, the manifest summary, and the reminder that every stub is empty. Open one stub if asked — show, do not narrate.

Human stop

The scaffold is structure, not content, and it is not authorization. Building anything inside it still requires the confirmed harness contract, and every semantic decision (Revenue included) stays with its named owner.