How to use Claude to implement Markdown tasks
The Backlog.md creator reported improving an agent coding workflow through project documentation, individual Markdown tasks, planning, and review before merging.
news.ycombinator.com
Prompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
For [project] and [feature specification], use [README.md] and [CLAUDE.md] as context. Generate separate Markdown task files, self-review them, and produce an implementation plan. Wait for my plan edits, then implement the agreed tasks for review before merging.
Reach people exploring AI agents
See placements →What happened
The author reported task success rising from 50 percent to 75 percent and then above 95 percent as the workflow changed; measurement details were not supplied.
Reported by mrlesk on Hacker NewsHow the agent did it
- Add README.md and CLAUDE.md for project context.
- Write a separate Markdown file for each task.
- Build Backlog.md to generate task files from specifications.
Where a person stepped in
The author tweaks plans when needed and describes reviewing and merging implemented changes.
A person approved each step before the agent continued.
The creator of Backlog.md described abandoning an early attempt to use Claude Code on an existing codebase because untangling its output took longer than writing from scratch. Their subsequent changes focused on the development process. They first added README.md and CLAUDE.md files so the model had project context, reporting task success of 50 percent at that stage.
Next, they wrote a separate Markdown file for each task, with Codex planning and Claude coding. They reported 75 percent success with that arrangement. They then built Backlog.md, a CLI that automatically turns a high-level specification into task files, using Claude and Codex to build the tool itself. The author associated that stage with task success above 95 percent; the post does not explain how those percentages were measured.
Their working loop starts with task generation by Codex or Claude Opus and self-review. The same agent then produces a plan in plan mode, which the author tweaks if needed. Claude Sonnet or Codex implements the work, followed by review and merging. Who it's for: developers who want explicit task files and a reviewable plan before agent implementation.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · mrlesk
Frequently asked questions
What project context did the author add?
They added README.md and CLAUDE.md so the model knew the project, then introduced one Markdown file per task.
What is the reported implementation loop?
Generate tasks and self-review them, generate a plan and tweak it if needed, then implement with Claude Sonnet or Codex before review and merge.
More details
Remaining steps
- Generate and self-review tasks, then generate a plan.
- Tweak the plan if needed, implement, review and merge.
About this account
Claude and Codex generate tasks, plans, and implementation within a staged workflow.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=44483531. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: mrlesk on Hacker News
Source published 6 Jul 2025
Source checked 14 Oct 2026
Tools and topics
Claude Code · Codex · Backlog.md · Markdown
AI coding
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.