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.

インストールへ移動

リポジトリ: luanmorenommaciel/uc-transact-co
ソースの最終更新活動: 2026年8月15日 00:51
検出された SKILL.md の言語: 英語
スター: 49
フォーク: 48

インストール方法

デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。

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

コマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。

ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。

ソースファイルを確認

インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。

ファイルエクスプローラー

2 ファイル

SKILL.md

ソースの指示 · 読み取り専用プレビュー

検索に戻るインストールに戻る

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

このリポジトリのすべてを見る

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.