Reusable AI development workflows
without building another runtime.

THE PROBLEM

AI coding tools are powerful, but long-running work easily loses structure.

Multi-step engineering work can drift across sessions, token limits and interruptions. Important requests get forgotten, architecture gets duplicated, unfinished work is abandoned, and deployment or security checks are often left until the end.

Open Tools addresses that by turning repeatable engineering behaviours into explicit agents, commands, skills and workflow packs rather than relying on a general-purpose assistant to remember everything.

WORKFLOW ARCHITECTURE

Four reusable primitives: agents, commands, skills and packs.

• Agents define focused AI roles with explicit purpose, permissions and operating rules.
• Commands expose those roles through repeatable task-specific entry points.
• Skills package reusable engineering guidance that multiple agents can load.
• Packs group commands into end-to-end workflows such as memory, preflight, feature delivery and AI product design.

The entire system is distributed as readable Markdown/YAML content, keeping the workflow portable and easy to customize without introducing a separate runtime or dependency layer.

PERSISTENT WORK MEMORY

An ADHD-friendly task ledger for interrupted, non-linear work.

• A canonical ADHD_TASKS.md ledger captures every non-trivial request as a visible checkbox.
• Later requests are appended without silently replacing the active task.
• Complex prompts are broken into explicit phases.
• Completed work is checked off only when evidence exists.
• Interruptions can resume from the active phase instead of restarting from scratch.
• Unrelated ideas are parked separately rather than derailing the current task.
• A compact session kanban provides machine-oriented recovery across sessions.

SOFTWARE DELIVERY AGENTS

Specialized roles for the full development lifecycle.

Open Tools includes dedicated agents for product strategy, backend, frontend, QA, deployment, security, SEO, UX and release management.

• Full-stack endpoint delivery with backend, frontend and contract wiring checks.
• No-drift feature building that verifies existing routes, services and data contracts before adding new surfaces.
• Repository mapping and idle-surface detection.
• Diff and PR review for regressions, missing tests and security issues.
• Release readiness checks across git state, tests, migrations, documentation, deployment and rollback planning.
• Railway-specific deployment and service health workflows.

AI PRODUCT DESIGN

Bounded AI workflows instead of unconstrained prompts.

• Agent workflow design with explicit inputs, outputs, validation and audit boundaries.
• Structured-output design for strict JSON schemas, retries and failure handling.
• Document extraction workflows that preserve source order, uncertainty and exact text where required.
• Controlled domain classification with labels, confidence and reasons.
• Evidence-auditability skills for legal, compliance and case-building systems.

SAFETY MODEL

Automation with explicit boundaries.

• Review and audit agents are read-only by default.
• Memory agents are restricted to controlled project locations unless explicitly authorized.
• Credentials, tokens, cookies, database URLs and private keys are excluded from persistent memory.
• Destructive Git commands and force pushes are prohibited by default.
• Publish and merge actions require explicit authorization and can include verification gates before completion.

DISTRIBUTION

Portable by design.

Open Tools ships with an installer, manifest, templates, documentation and an MIT license. Teams can copy the workflow pack directly into a project, customize individual agents or skills, and keep the resulting behaviour version-controlled alongside the codebase.

WHAT THIS DEMONSTRATES

Agent design, developer experience and workflow engineering.

Open Tools demonstrates the ability to design practical multi-agent systems around real software delivery: persistent context, permissions, testing, architecture discipline, release checks, deployment verification, security boundaries and human-controlled automation.