skills.rest
curdx
curdxCOMMUNITY
Explore all skills by this author →
GitHub

implement

Execute spec-driven tasks from tasks.md via a coordinator loop with verification gates.

4|Updated Apr 26, 2026
Search Tags:#state management#verification#Claude Code#task delegation#spec execution#autonomous tasks#cost runaway guard
One-click install
npx skills add https://github.com/curdx/curdx-flow --skill implement-curdx
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: implement
Source: https://github.com/curdx/curdx-flow/tree/main/plugins/curdx-flow/skills/implement
Command: npx skills add https://github.com/curdx/curdx-flow --skill implement-curdx

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It solves the problem of getting Claude Code to reliably execute a spec’s tasks without context rot, silent failures, or runaway iteration during implementation.

Core Features & Use Cases

  • Goal-driven autonomous execution: Uses the spec’s tasks.md to drive a coordinator loop and delegates implementation via Claude Code agents (including VERIFY tasks).
  • Resumable, audit-friendly state: Initializes and merges a deterministic .curdx-state.json execution record so progress survives restarts.
  • Verification and completion gates: Requires verification layers and evidence protocols before emitting ALL_TASKS_COMPLETE.
  • Safety controls for cost runaway: Enforces --max-task-iterations and --max-global-iterations caps, halting deterministically when exceeded.
  • Failure recovery option: When --recovery-mode is enabled, failed tasks can trigger automatic fix task generation within configured limits.

Quick Start

Use the implement skill to execute the current active spec tasks by running: npx @curdx/flow implement <your-arguments>

Frequently Asked Questions about implement

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

FAQPage Schema
How do I automate spec-driven task execution in Claude Code without context rot?▼

Automated spec-driven task execution uses a coordinator loop to delegate work from tasks.md to Claude Code agents. It initializes a deterministic state record to maintain progress, preventing context rot and silent failures across implementation cycles.

How do I prevent cost runaway when running autonomous implementation tasks?▼

To prevent cost runaway during autonomous implementation tasks, enforce iteration caps using max-task-iterations and max-global-iterations limits. The coordinator loop halts deterministically when these safety thresholds are exceeded.

Can I resume an interrupted autonomous coding workflow if the process stops?▼

Yes, autonomous coding workflows are resumable because the system initializes and merges a deterministic .curdx-state.json execution record. This audit-friendly state file ensures progress survives restarts without data loss.

How does verification work for autonomous task delegation to prevent completion hallucinations?▼

Verification for autonomous task delegation requires dedicated verification layers and evidence protocols before emitting an ALL_TASKS_COMPLETE signal. This gating mechanism prevents completion hallucinations by demanding proof of successful implementation.

How do I automatically generate fix tasks for failed autonomous implementation steps?▼

Enable recovery-mode to automatically generate fix tasks for failed autonomous implementation steps. This failure recovery option triggers automatic fix task generation within configured iteration limits to resolve execution errors.

Do I need a specific prerequisite setup to run autonomous spec execution?▼

Yes, autonomous spec execution requires prerequisite validation and state initialization with merged execution fields. An active curdx-flow spec containing a tasks.md file must be present before running the implementation workflow.

Related Skills

Explore More Software Engineering Skills→
curdxcurdx
4

curdx-core

Coordinate curdx-flow execution by governing flags, state lifecycle, and subagent delegation.

Community
Advanced
benjaminshoemakerbenjaminshoemaker
39

populate-state

Generate .claude/phase-state.json from EXECUTION_PLAN.md and git history.

Community
Intermediate
bennybennisonbennybennison

autonomous-loops

Execute autonomous agent loops with stop conditions and iteration caps.

Community
Advanced
dawsonblockdawsonblock

subagent-driven-development

Delegate independent tasks to fresh subagents with two-stage spec and quality reviews.

Community
Advanced