AI Dev Workflow Kit — Summary

AI Dev Workflow Kit by aahilsayed062 is a portable .ai-kit/ folder drop-in framework enforcing the EPCC Engine (Explore → Plan → Code → Verify → Audit) on any AI coding agent. Its core constraints are: Max-2 Rule (at most 1 PRIMARY + 1 SUPPORT resource active simultaneously), mandatory Decision Gates A–H (forcing the agent to ask questions and lock decisions before coding on choice-heavy tasks), a Status Banner printed at the start of every response, and Start Menu templates that boot the agent with the correct resource pair. Resources are markdown "knowledge cards" that constrain agent behavior per task type (technical-cofounder-contract, cinematic-landing-lane, ui-inspiration-to-lane, etc.). The framework ships 9 shipped resource cards plus 12 template combinations (I–L are default/shipped, A–H are legacy/extension). No runtime required — pure markdown instruction files. 4 stars, MIT license, last commit 2026-03-01.

Differs from seeds: Closest to agent-os (Archetype 4 — markdown scaffold) but with a more opinionated constraint system. Unlike agent-os (which writes the standards once and defers to the agent), ai-dev-workflow-kit enforces structural compliance per-response via the Status Banner and per-task via Decision Gates. The Max-2 Resource Rule is a novel context-budgeting discipline not seen in any seed. The resource-card / template-boot model is closer to bmad-method's persona files but more restrictive (max 2 resources vs BMAD's full persona stack).

AI Dev Workflow Kit — Overview

Origin

By aahilsayed062. 4 stars, MIT license, last commit 2026-03-01. Has AGENTS.md, CHANGELOG.md, ROADMAP.md — suggesting planned multi-agent evolution.

Mission (verbatim)

"AI Dev Workflow Kit is a portable folder (.ai-kit/) you drop into any project to give your AI coding agent: EPCC Engine — a mandatory workflow loop: Explore → Plan → Code → Verify → Audit; Resource Cards — reusable knowledge cards that constrain agent behavior per task; Decision Gates — force the agent to ask questions, propose options, and lock decisions before coding on choice-heavy tasks; Max-2 Rule — at most 2 active resources (1 PRIMARY + 1 SUPPORT) to prevent context overload."

Core principles

Escape hatch

"Exact phrase only: 'INPUTS ARE FIXED, SKIP GATES.'" — the only documented override to bypass Decision Gates.

Design philosophy

Novel context-budgeting discipline: preventing context overload by limiting active resources to 2. This is a direct engineering response to the "token-intensive framework" failure mode cited in Reddit critiques of superpowers/BMAD.

AI Dev Workflow Kit — Architecture

Distribution

standalone-repo — copy .ai-kit/ directory into any project.

Install

# Clone, then install into your project
./scripts/install.sh /path/to/your/project

# Or manually
cp -R .ai-kit/ /path/to/your/project/.ai-kit/

Directory structure

.
├── .ai-kit/
│   ├── AGENTS.md            (multi-agent spec)
│   ├── CANONICAL.md         (version authority + manifest)
│   ├── CURRENT-VIEW.md      (current active state)
│   ├── README.md
│   ├── VERSIONING.md
│   ├── prompts/
│   │   └── START.md         (template boot menu)
│   └── resources/           (9 shipped knowledge cards)
│       ├── anti-ai-slop-details.md
│       ├── cinematic-landing-lane.md
│       ├── component-logic-spec.md
│       ├── content-architecture.md
│       ├── deliverables-specs.md
│       ├── landing-bg-asset-pipeline.md
│       ├── reference-style-extraction.md
│       ├── technical-cofounder-contract.md
│       └── ui-inspiration-to-lane.md
├── AGENTS.md                (root-level agent spec)
├── scripts/
│   └── install.sh
└── examples/

Runtime requirements

None. Pure markdown instruction files.

Target AI tools

Any AI coding agent (Claude Code, Cursor, Copilot, etc.) — designed for maximum portability.

Version authority

.ai-kit/CANONICAL.md is the single source of truth for KIT_VERSION (currently v2.0.1).

AI Dev Workflow Kit — Components

No runnable Claude Code primitives

0 commands (.claude/commands/), 0 skills, 0 agents, 0 hooks. Framework operates via markdown instruction files that the agent reads.

Resource Cards (9 shipped in .ai-kit/resources/)

Template Combinations (12 in START.md)

Default shipped (I–L)

Legacy/extension (A–H, user provides cards)

A: Universal Boot, B: Backend SaaS, C: Dashboard Feature, D: Landing Page, E: UI Polish, F: Reliability Pass, G: Debug, H: Ship-Ready QA

Canonical Commands (from CANONICAL.md)

These are text commands typed in conversation, not slash commands:

• SHOW_ACTIVE_RESOURCES
• RUN_DECISION_GATES
• SHOW_DECISION_STATE
• GENERATE_DECISION_REPORT
• LOCK_DECISIONS
• COFOUNDER_MODE

Decision Gates (A–H)

8-gate question system for choice-heavy decisions. Max 8 questions per round. Escape: "INPUTS ARE FIXED, SKIP GATES."

AI Dev Workflow Kit — Uniqueness

Differs from seeds

Closest to agent-os (Archetype 4 — markdown scaffold), but with a novel context-budgeting discipline (Max-2 Rule) not present in any seed. Unlike superpowers or bmad-method, which inject behavioral constraints via Iron Laws or persona prompts and leave context management to the agent, ai-dev-workflow-kit explicitly limits active resources to 2 to prevent context overload. The Decision Gate system (A–H, exact-phrase escape) is more structured than any seed's approval gate model. The CANONICAL.md single-source-of-truth pattern is a versioning discipline not seen in seeds. The EPCC framework is similar in spirit to spec-kit's phased workflow but more generic (not spec-document-specific).

Positioning

A lightweight, tool-agnostic workflow enforcement kit for developers who want structured AI behavior without vendor lock-in or complex tooling. Particularly targeted at small teams or solo developers building landing pages and SaaS products (the shipped templates are UI-heavy).

Observable failure modes

1. Instruction-level only: "Rules are instruction-level constraints; enforcement depends on agent compliance, but the kit is designed to make drift visible and hard to justify" — acknowledges the fundamental limitation.
2. Templates are UI-focused: The 4 shipped templates (I–L) are all landing-page/marketing-site oriented; general software development requires user-provided cards.
3. 4 stars: Minimal adoption; largely untested in production environments.
4. Status Banner compliance: The Status Banner depends on the agent consistently including it; there's no hook or technical enforcement.
5. Decision Gate skip abuse: The "INPUTS ARE FIXED, SKIP GATES." escape requires agent recognition of the exact phrase — not technically enforced.