next

Orchestrates one unit of work through enforced plan, analyze, approval, execute, and review phase gates.

Updated Jun 28, 2026
One-click install
npx skills add https://github.com/sicambria/talkteach-asr --skill next-sicambria
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: next
Source: https://github.com/sicambria/talkteach-asr/tree/main/.claude/skills/next
Command: npx skills add https://github.com/sicambria/talkteach-asr --skill next-sicambria

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Non-trivial code changes often ship without a grounded plan, skip human approval, or bypass verification because process checklists are advisory and easy to ignore. This Skill enforces a hard state machine — plan → Phase-1 gate → analyze → human approval → execute → verify/review — where every transition is blocked by a gate script if the required artifact or approval is missing. ## Core Features & Use Cases - Enforced phase gating: Each transition calls next-phase-gate.mjs advance <phase>, which exits 1 and blocks out-of-order jumps or missing precondition artifacts (plan file, analyze report, recorded approval). - Fresh-context executor subagents: Implementation runs in isolated git worktrees via executor subagents that return BRANCH READY and never self-merge; the orchestrator integrates serially with --no-ff and re-verification between merges. - Mandatory adversarial review: Verifier, slopchecker, and adversarial-reviewer subagents must all pass per branch, and a review that finds zero issues is treated as incomplete and halts the flow. - Use Case: You need to land a risky feature in a repository with strict quality standards. Run /next to force a grounded plan with standards evidence, a cross-artifact consistency analysis, explicit human sign-off, and gated execution with adversarial review before any merge. ## Quick Start Run the /next command to start a new gated unit of work, beginning with a reset and a plan containing a Standards & Guardrails Evidence section.

Frequently Asked Questions about next

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

FAQPage Schema
How do I run a gated plan-to-execution workflow for a code change?▼

Invoke /next, then reset the gate state and write a plan with a Standards & Guardrails Evidence section. Advance through phase1, analyze, approval, and execute in order — the next-phase-gate.mjs script blocks any transition whose precondition artifact is missing.

What is the difference between /next and agent-protocol?▼

/next drives one unit of work linearly through hard phase gates with a human approval in the middle. agent-protocol fans a fleet of agents across an independent backlog in parallel waves with no per-item approval gate, relying on serialized integration and adversarial verification instead.

Why does the execute phase get blocked after planning?▼

The gate blocks execute unless the analyze report exists on disk and a human approval has been recorded via the approve command. Advance approval with the analyze report path, then run approve --by <human> before advancing to execute.

Can executor subagents merge their own branches?▼

No. Executors work in isolated git worktrees and return BRANCH READY without merging. The orchestrator integrates branches serially with --no-ff merges, re-running verification between each merge.

What happens if the adversarial review finds no issues?▼

A review with zero findings is treated as incomplete and halts the flow rather than passing. The adversarial-reviewer is a must-find-issues edge-case hunter, so an empty result signals insufficient scrutiny.