bounded-work-item

Executes one scoped implementation item and returns verification evidence to the assigned reviewer.

3.5k|342|Updated Oct 25, 2025
One-click install
npx skills add https://github.com/bfly123/claude_code_bridge --skill bounded-work-item
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: bounded-work-item
Source: https://github.com/bfly123/claude_code_bridge/tree/main/docs/plantree/plans/agentic-loop-workflow/drafts/agentroles.coder/skills/bounded-work-item
Command: npx skills add https://github.com/bfly123/claude_code_bridge --skill bounded-work-item

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

In multi-agent coding workflows, agents often drift beyond their assigned scope, touch unrelated files, or claim unverified success. This Skill constrains a Coder agent to exactly one canonical work packet, enforcing strict boundaries on what can be read, edited, and submitted.

Core Features & Use Cases

  • Scoped Execution: Reads only the declared work packet and its refs, then makes the smallest change satisfying the packet inside allowed paths.
  • Evidence-Based Submission: Runs focused verification and submits results to the assigned Reviewer through the assigned-review chain, returning a terminal status of done, blocked, or needs_rework.
  • Boundary Enforcement: Prohibits editing authoritative state, running workflow wrappers, expanding scope, or claiming whole-round success.
  • Use Case: In an agentic loop workflow, a controller dispatches a node work packet to a Coder agent; the agent implements the change, verifies it, submits evidence for review, applies bounded rework if requested, and stops.

Quick Start

Execute this single work packet within its declared scope and allowed paths, verify the change, and submit the evidence to the assigned reviewer.

Frequently Asked Questions about bounded-work-item

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

FAQPage Schema
How do I execute a bounded work item in an agentic coding workflow?

Read only the canonical node work packet and its declared refs, inspect relevant files, make the smallest change inside the allowed paths, run focused verification, then submit evidence to the assigned Reviewer through the review chain and stop.

What terminal statuses can a coder agent return for a work item?

The only valid terminal statuses are done, blocked, and needs_rework. After Reviewer pass, the agent returns changed paths, verification evidence, blockers, and done without further file or tool mutation.

What happens if the assigned workspace is not a Git checkout?

Detect repository metadata once and stop retrying Git commands if the workspace is not a checkout. Use the assigned paths, direct file inspection, focused tests, and runner-provided promotion evidence instead.

What actions are forbidden when executing a bounded work item?

The agent must not lower acceptance criteria, expand scope, edit authoritative state or runtime files, run workflow wrappers, submit downstream asks, create authority commits, integrate sibling work, or release agents.

When should I not use a bounded work item approach?

Avoid it when the task requires cross-packet integration, whole-round coordination, or direct state promotion, since those responsibilities belong to the controller and reviewer roles rather than the coder.