skills.rest
longdang193
longdang193COMMUNITY
Explore all skills by this author →
GitHub

skill-deepagents-executing-plans

Executes approved Git-tracked implementation plans through bounded DeepAgents delegation via dcode-project.

Updated Apr 9, 2026
Search Tags:#git workflow#codex#multi-agent#plan execution#task delegation#DeepAgents#evidence-based acceptance
One-click install
npx skills add https://github.com/longdang193/project-OS-starter --skill skill-deepagents-executing-plans-longdang193
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: skill-deepagents-executing-plans
Source: https://github.com/longdang193/project-OS-starter/tree/main/generated_agents/claude/skills/skill-deepagents-executing-plans
Command: npx skills add https://github.com/longdang193/project-OS-starter --skill skill-deepagents-executing-plans-longdang193

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Coordinating multi-agent code execution is risky: agents may exceed their scope, lose state on failure, or claim success without evidence. This Skill governs how a Codex lead controller dispatches bounded DeepAgents tasks against an approved Git-tracked plan, ensuring every result is reconciled with Git and plan state before acceptance. ## Core Features & Use Cases - Bounded Dispatch Contracts: Each dispatch is tied to exactly one active plan task with explicit workspace identity, write ownership, dependencies, and required proof. - MCP Escalation Contract: Controls when DeepAgents may use MCP capabilities, returning NEEDS_CONTEXT when live data is missing instead of improvising. - Evidence-Based Acceptance: Runtime success never completes a plan task; Codex inspects Git state, diffs, and proof before updating the task ledger. - Use Case: A lead controller has an approved multi-task implementation plan. It dispatches one dependency-ready task to DeepAgents, receives a DONE_WITH_CONCERNS claim, reconciles the diff against declared scope, reruns required proof, and only then advances the plan ledger. ## Quick Start Ask the agent to execute the next dependency-ready task from the approved plan using bounded DeepAgents delegation with evidence-backed acceptance.

Frequently Asked Questions about skill-deepagents-executing-plans

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

FAQPage Schema
How do I execute an implementation plan with DeepAgents sub-agents?▼

Dispatch one bounded task at a time through dcode-project with an explicit brief covering workspace identity, scope, write ownership, dependencies, and required proof. The lead controller reconciles each returned claim against Git and plan state before advancing the task ledger.

What is the difference between DeepAgents runtime state and plan coordination state?▼

DeepAgents runtime state and internal task decomposition are ephemeral and executor-local. The Git-tracked plan and its task ledger are the only durable coordination state, written solely by the lead controller after evidence-backed acceptance.

Can DeepAgents use MCP tools during plan execution?▼

DeepAgents may use MCP only when the launcher receives explicit --mcp-select selectors for approved servers; otherwise it runs with --no-mcp. If mid-task data is missing or stale, it returns NEEDS_CONTEXT rather than improvising access.

When should parallel DeepAgents writers be used?▼

Parallel writers require dependency-ready plan tasks, isolated Git worktrees, and disjoint write ownership. Concurrency permission comes from plan dependencies plus Git isolation, never from runtime concurrency capability alone.

What happens when a DeepAgents task returns BLOCKED or FAIL?▼

Acceptance stops immediately and plan task state does not advance. The controller preserves safe partial changes, records blocker evidence under the active task, and reconciles Plan plus Git before any retry, debugging, or escalation.

Related Skills

Explore More Software Engineering Skills→
alibabaalibaba
5.5k

teamharness-team-coordination

Plans task boundaries, ownership, and DAG or Loop execution modes for multi-agent team coordination.

Official
Intermediate
xaionaroxaionaro

finishing-a-development-branch

Verify passing tests and present merge, PR, keep, or discard options for Git branches.

Community
Intermediate
chenziyang110chenziyang110
7

Subagent-Driven Development (Spec Kit Plus)

Route delegated work through sp-* workflows with validated WorkerTaskPacket contracts.

Community
Intermediate
t-udat-uda

light-orchestration

Determine whether to execute a task directly or decompose it into a bounded multi-agent plan.

Community
Intermediate