do-next

Dispatches NEXT.md open items as worktree-isolated sub-agents and merges each landed track.

Updated Aug 4, 2026
One-click install
npx skills add https://github.com/polycode-public/the-mechanical-code-talker --skill do-next-polycode-public
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: do-next
Source: https://github.com/polycode-public/the-mechanical-code-talker/tree/main/.claude/skills/do-next
Command: npx skills add https://github.com/polycode-public/the-mechanical-code-talker --skill do-next-polycode-public

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating a backlog of open engineering tasks across parallel agents is error-prone: merge collisions on shared files, stale tracking notes, unlanded worktrees, and bugs silently dropped. This Skill turns a repository's NEXT.md backlog into a disciplined dispatch-and-land loop where each work item is executed in an isolated git worktree and merged as soon as it is ready. ## Core Features & Use Cases - Backlog-to-batch dispatch: Reads NEXT.md's Open items, filters out non-actionable notes, and groups items by file ownership into conflict-free tracks. - Worktree-isolated sub-agents: Each track runs in its own git worktree with a self-contained brief covering file ownership, minimal test commands, git discipline, and a report-back contract. - Continuous landing and tracking: Merges, tests, and pushes each track the moment it completes, updates NEXT.md's in-flight block in the same breath, and refills freed files with the next queued items. - Use Case: A maintainer returns to a repo with eight open items in NEXT.md, says "do next", and the session dispatches three parallel tracks, lands each as it finishes, records surfaced bugs as sub-clauses, and rolls the patch version once CI is green. ## Quick Start Tell the agent to follow the do-next skill and work the open items in NEXT.md.

Frequently Asked Questions about do-next

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run a backlog of tasks with parallel AI agents?▼

Keep open items in a NEXT.md file and invoke this skill to dispatch them as worktree-isolated sub-agents. Items are grouped by file ownership so parallel tracks never edit the same file, and each track is merged, tested, and pushed as soon as it lands.

How do I avoid merge conflicts when multiple agents edit one repository?▼

Assign each agent an isolated git worktree plus an explicit file-ownership list in its brief. Tracks that must touch the same file are sequenced or scoped to non-overlapping regions, and only the coordinator merges to main.

Can sub-agents push or merge to main directly?▼

No. Sub-agents commit inside their own worktree only; pushing, merging to main, and editing NEXT.md are reserved for the coordinator. This prevents concurrent edits to the tracking file and unreviewed changes reaching main.

What happens when a sub-agent finds an unrelated bug?▼

The bug is recorded in NEXT.md as a sub-clause of the item that surfaced it, keeping that item open, unless it is genuinely unrelated to the item's scope. Failing tests a track reports become work for the current batch rather than being carried forward.

When should I not dispatch a backlog item as a track?▼

Do not dispatch items that are documentation notes or gotcha reminders rather than actionable changes. Also avoid manufacturing unrequested work to keep agents busy; when NEXT.md is empty or fully claimed, the correct state is to stop.