← ClaudeAtlas

dandorilisted

Author a surgical agent task spec—a structured markdown file an autonomous coding agent reads to execute one bounded unit of work. Use when the user wants to draft an agent task, write a spec for an agent or Claude Code, package a conversation as something an agent can run independently, turn planning decisions into an executable instruction, or convert a discussion into a buildable task. Also trigger on phrases like "write me a prompt for the agent," "package this for Claude Code," "make this into a task spec," "turn this into something agentic to run," or "draft an instruction the agent can execute." Produces a markdown file with goal, files, required changes, acceptance criteria, verification commands, and commit format. Handles standalone tasks and tasks descending from a larger project document. Interrogates only gaps it can't infer from context; never asks for what's already visible.
sageframe-no-kaji/dandori · ★ 0 · Data & Documents · score 70
Install: claude install-skill sageframe-no-kaji/dandori
# Dandori (段取り) Dandori is the practitioner's prep discipline—the carpenter's setup, the chef's mise en place. This skill does the dandori for an autonomous coding agent: it translates a conversation's decisions into a surgical task spec the agent can execute independently. ## What this skill produces A single markdown file in `agent-tasks/`, named `agent-task-YYYY-MM-DD-slug.md`, containing the bounded spec for one unit of work. The agent reads it and produces the output, verifying against explicit acceptance criteria. The format is documented in `FORMAT.md`. The output is a structured spec, not a prompt. It exists as an artifact in the project so the agent can read it directly and the practitioner has a record of what was authorized. ## When this skill is in scope Right move when: - The conversation has produced enough (clear goal, file boundaries, definition of done) to bound a unit of agent work. - The practitioner wants a surgical handoff to an autonomous agent (Claude Code, Codex, Cursor agent mode, or equivalent), not an exploratory chat. - A standalone maintenance or exploration task needs authoring independently of any project context. Not the right move when: - The work is conversational—small refactors, single-line changes, debugging-by-discussion. Just do it. - The conversation hasn't produced architectural decisions yet. Have the design conversation first; author the task after. - The work is too large for a single bounded session. Decompose first; autho