skills.rest
yjwipod-1
yjwipod-1COMMUNITY
Explore all skills by this author →
GitHub

railyard

Orchestrates role-based agent workflows with SQLite-backed tickets, epics, and deterministic validation gates.

6|Updated Apr 25, 2026
Search Tags:#agent workflow#multi-agent#orchestration#sqlite#validation contract#task state#ticket lifecycle
One-click install
npx skills add https://github.com/yjwipod-1/railyard --skill railyard-yjwipod-1
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: railyard
Source: https://github.com/yjwipod-1/railyard
Command: npx skills add https://github.com/yjwipod-1/railyard --skill railyard-yjwipod-1

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires jsonschema, referencing, mcp, and includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve? Long-running AI-agent projects lose state when task tracking lives only in chat history, and execution quality suffers when planning, implementation, and review share one session. Railyard separates Planner, Architect, Runner, and Validator roles with durable SQLite workflow state so work survives across disposable sessions. ## Core Features & Use Cases - Role-based workflow protocol: Human, Planner, Architect, Runner, and Validator roles with explicit lane separation (Domain and System), ticket lifecycle rules, and closed-loop Architect review of Runner results. - Durable SQLite state: Epics, tickets, claims, results, and reviews persist in a local workflow database, with helper scripts for dispatch, claim, stale-ticket recovery, and review recording. - Deterministic validation: Validation contracts, a Validator gate on tickets, a failure taxonomy for blockers, and a v0.8 local runtime with event journal, gate decisions, action policy, and a 20-scenario smoke suite. - Use Case: A team running multi-session agent development clones Railyard into their project, initializes the workflow database, and has a Planner decompose work into epics and tickets that Architects dispatch to Runner sessions with independent Validator evidence before acceptance. ## Quick Start Clone the repository into your project as a railyard subdirectory, run python railyard/scripts/init_workflow.py --project-root ., then start a Planner session asking it to read railyard/SKILL.md and convert your project direction into epics and tickets.

Frequently Asked Questions about railyard

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

FAQPage Schema
How do I set up a multi-agent workflow with persistent task state?▼

Clone Railyard into your project as a railyard/ subdirectory and run python railyard/scripts/init_workflow.py --project-root . to create the SQLite workflow database, agent profiles, and mailbox directories. Then start Planner, Architect, and Runner sessions using the provided startup prompts.

How do I keep AI agent task state across separate chat sessions?▼

Store workflow state in a SQLite database rather than conversation history. Railyard records epics, tickets, claims, results, and reviews in .workflow/workflow.db, so each new session reads the database and resumes from durable state.

Does Railyard require specific AI models or hosted services?▼

No. Railyard is platform-neutral and uses only the Python 3.10+ standard library for workflow helpers and runtime components. It does not include LLM API calls, model routing, hosted orchestration, or a scheduler.

What happens when a Runner agent gets blocked on a ticket?▼

The Runner reports a blocked result using the failure taxonomy: permission_denied, command_failed, sandbox_boundary, authorization_required, environment_issue, or unresolved_dependency. After three same-kind failures, the workflow stops and reports a structured blocker instead of retrying indefinitely.

When is independent Validator evidence required for a ticket?▼

The Planner or Architect records an explicit validator gate decision in each ticket's metadata. Tickets involving data transforms, migrations, source-to-derived artifacts, or high-risk implementation require Validator evidence, and Architect acceptance is prohibited until a pass report is verified.

Related Skills

Explore More Software Engineering Skills→
AidenSbVevoAidenSbVevo
1

epic-planning

Decompose project epics into PR-sized Linear issues with acceptance criteria.

Community
Advanced
prompt-securityprompt-security
1.1k

picoclaw-security-guardian

Detects configuration drift, filters security advisories, and verifies signed release artifacts for Picoclaw gateways.

Official
Advanced
volter-aivolter-ai

develop

Implements GitHub issue acceptance criteria, commits evidence, and pushes a branch for PR review.

Official
Advanced
hadi-technologyhadi-technology
1

phased-workflow

Orchestrate multi-stage software delivery with planning, review, implementation, and verification phases.

Official
Advanced