Skip to content
/

Agent OS

agent-os · buildermethods/agent-os · ★ 4.6k · last commit 2026-05-05

Extracts implicit codebase conventions into token-efficient markdown standards files and injects them selectively into AI agent context windows so agents build the way your team actually builds.

Best whenStandards documentation — not orchestration or scaffolding — is the single highest-leverage investment in AI-assisted development, and AI tools should be tru…
Skip ifBloated, prose-heavy standards files, Loading all standards into context on every session
Primitive shape 5 total
Commands 5
00

Summary

Agent OS — Summary

Agent OS is a lightweight, shell-installed framework that extracts coding standards from your codebase, stores them as markdown files, and injects them into AI agent context windows on demand. It solves the "standards drift" problem where AI coding agents ignore or forget project conventions because those conventions were never written down and made discoverable. What makes Agent OS distinct is its three-layer model — global tech-stack profile, per-project standards documents, and timestamped spec folders — which separates how you build (standards) from what you're building (specs), letting the AI reference both with surgical precision rather than flooding context with irrelevant instructions. It is designed for professional developers and small teams already using Claude Code, Cursor, or similar AI coding tools who want consistent, standards-aware code generation without heavy orchestration overhead. Agent OS is production-ready: it is at version 3.0 (released January 2026), has over 4,600 GitHub stars, 740 forks, 18 contributors, and a maintained changelog.

01

Overview

Agent OS — Overview

Origin and Author

Agent OS was created by Brian Casel of Builder Methods, a developer education brand focused on professional AI-first software development. The initial commit was made July 16, 2025. Version 3.0 — a major architectural rethink — shipped January 20, 2026.

Tagline

"Agent OS is a system for injecting your codebase standards and writing better specs for spec-driven development."

— GitHub repository description

The marketing site extends this to:

"Agents that build the way you would."

Philosophy and Problem Framing

Brian Casel built Agent OS to address a fundamental mismatch: AI coding agents are powerful but context-blind. They don't know that your team uses cursor-based pagination, error code envelopes like AUTH_001, or that migrations must always include a rollback. Developers encode this "tribal knowledge" in their heads, not in files — and every new session, every new subagent, starts from scratch.

The original v1 release (July 2025) tackled this with a multi-agent orchestration model that scripted out plan, spec, task, and implementation phases. By v3 (January 2026), Brian recognized that AI tools like Claude Code had evolved: plan mode, todo tracking, and frontier model reasoning now handle the orchestration naturally. Rather than fight that trend, he simplified Agent OS to do what it does uniquely well:

  1. Discover — Interactively surface implicit patterns from existing code into documented standards.
  2. Store — Keep those standards as minimal-token, scannable markdown files in agent-os/standards/.
  3. Inject — Pull only the relevant standards into context at the moment of need, using an index for fast matching.
  4. Shape — Wrap Claude Code's plan mode with a structured shaping workflow that seeds the plan with product context and standards.

The CHANGELOG captures this philosophy shift explicitly:

"AI coding tools have evolved significantly since Agent OS's original release in mid-2025. Claude Code's plan mode, extended thinking, and improved models now handle much of the scaffolding that earlier versions provided... Rather than reinvent these functions, v3 focuses on Agent OS's core strengths: establishing standards, injecting them smartly, and enhancing spec-driven development."

What Agent OS Believes

  • Standards documentation should be a first-class artifact of every codebase, not an afterthought.
  • Smaller, scannable standards outperform long architecture documents — every word injected into a context window costs tokens.
  • Specs should follow a dated, discoverable folder convention (YYYY-MM-DD-HHMM-feature-slug/) so future sessions can find and re-use prior shaping work.
  • AI tools should defer to the developer's taste; Agent OS is a taste-encoding layer, not an autonomous decision-maker.

Version History Highlights

  • v1 (Jul 2025) — Initial release, multi-phase multi-agent orchestration.
  • v2.1 (Oct 2025) — Added Claude Code Skills support; introduced per-phase commands (shape-spec, write-spec, create-tasks, implement-tasks, orchestrate-tasks).
  • v3.0 (Jan 2026) — Removed implementation/orchestration phases (deferred to native AI tools); introduced /discover-standards, /inject-standards, index.yml-based smart matching; simplified profile inheritance to a single config.yml.
02

Architecture

Agent OS — Architecture

Distribution Type

bash-script-bundle

Agent OS is a standalone shell-script system cloned from GitHub. There is no npm package, no pip install, no plugin registry entry. The repo is cloned locally and a project-install script copies the right files into each project.

Install Method

Step 1: Clone the base installation

git clone https://github.com/buildermethods/agent-os.git ~/agent-os

Step 2: Install into a project (run from the project root)

cd /path/to/your/project
~/agent-os/scripts/project-install.sh

Optional flags:

~/agent-os/scripts/project-install.sh --profile rails          # Use a named profile
~/agent-os/scripts/project-install.sh --commands-only          # Only update commands, keep existing standards
~/agent-os/scripts/project-install.sh --verbose                # Show detailed output

What the install script does

  1. Resolves the active profile (default: default) from config.yml.
  2. Builds the profile inheritance chain (profiles can inherit from other profiles).
  3. Copies standards files from profiles/<profile>/standards/ to <project>/agent-os/standards/.
  4. Builds agent-os/standards/index.yml — a YAML index of all standards files with descriptions.
  5. Copies command markdown files to <project>/.claude/commands/agent-os/.

Claude Code vs Cursor install path differences

Tool Commands land at Standards land at
Claude Code .claude/commands/agent-os/ agent-os/standards/
Cursor .cursor/rules/ or manual injection agent-os/standards/ (same)

The README directs users to buildermethods.com/agent-os for full documentation covering Cursor-specific setup. Standards storage is identical for both tools; the difference is only where command markdown files are placed.

Syncing standards back to a profile

~/agent-os/scripts/sync-to-profile.sh
~/agent-os/scripts/sync-to-profile.sh --profile rails
~/agent-os/scripts/sync-to-profile.sh --new-profile nextjs --all

File / Directory Layout

Base installation (cloned repo at ~/agent-os/)

agent-os/
├── .github/
├── commands/
│   └── agent-os/
│       ├── discover-standards.md
│       ├── index-standards.md
│       ├── inject-standards.md
│       ├── plan-product.md
│       └── shape-spec.md
├── profiles/
│   └── default/
│       └── global/
│           └── tech-stack.md
├── scripts/
│   ├── common-functions.sh
│   ├── project-install.sh
│   └── sync-to-profile.sh
├── .gitignore
├── CHANGELOG.md
├── LICENSE
├── README.md
└── config.yml

Per-project layout (after install into <project>/)

<project>/
├── .claude/
│   └── commands/
│       └── agent-os/
│           ├── discover-standards.md
│           ├── index-standards.md
│           ├── inject-standards.md
│           ├── plan-product.md
│           └── shape-spec.md
└── agent-os/
    ├── product/
    │   ├── mission.md
    │   ├── roadmap.md
    │   └── tech-stack.md
    ├── specs/
    │   └── YYYY-MM-DD-HHMM-feature-slug/
    │       ├── plan.md
    │       ├── shape.md
    │       ├── standards.md
    │       ├── references.md
    │       └── visuals/
    └── standards/
        ├── index.yml           ← auto-maintained discovery index
        ├── global/
        │   └── tech-stack.md
        ├── api/
        │   └── *.md
        ├── database/
        │   └── *.md
        └── [other-folders]/

Required Dependencies

Dependency Required Notes
bash (or zsh) Yes Standard macOS/Linux shell
git Yes For cloning the repo
Claude Code Recommended Commands installed to .claude/commands/
Cursor Optional Manual command integration needed
python3 / node No None required

The scripts use only POSIX shell utilities (awk, find, cp, mv). A recent fix (PR #327) replaced GNU tac with POSIX awk for macOS compatibility.

Configuration Files

File Purpose
~/agent-os/config.yml Global default profile and profile inheritance definitions
<project>/agent-os/standards/index.yml Auto-generated discovery index for standards
<project>/agent-os/product/mission.md Product mission (created by /plan-product)
<project>/agent-os/product/roadmap.md Product roadmap (created by /plan-product)
<project>/agent-os/product/tech-stack.md Tech stack reference (created by /plan-product)

There is no CLAUDE.md at the project root. Standards are injected on demand via the /inject-standards command rather than always-loaded via CLAUDE.md.

Profile System

Profiles live in profiles/<name>/ inside the cloned repo. The default profile ships with a single standard (global/tech-stack.md). Profile inheritance chains are defined in config.yml:

version: 3.0
default_profile: default

# profiles:
#   profile-a:
#     inherits_from: default
#   profile-b:
#     inherits_from: profile-a

Custom profiles (e.g., rails, nextjs) can be created and shared via the sync script.

03

Components

Agent OS — Components

Commands

Agent OS ships 5 slash-commands as markdown files installed to .claude/commands/agent-os/.

Command File Purpose
/discover-standards discover-standards.md Interactively analyze the codebase, surface implicit patterns, and create documented standards files via a Q&A workflow with the user
/inject-standards inject-standards.md Inject relevant standards into current context; supports auto-suggest mode (AI picks based on task) and explicit mode (/inject-standards api/response-format)
/index-standards index-standards.md Rebuild agent-os/standards/index.yml — scan all standards files, propose one-line descriptions, prune stale entries
/plan-product plan-product.md Interactive product documentation wizard: creates mission.md, roadmap.md, and tech-stack.md in agent-os/product/
/shape-spec shape-spec.md Spec-shaping workflow that runs inside Claude Code plan mode; gathers visuals, references, product context, and relevant standards, then produces a timestamped spec folder

Skills

(none)

Agent OS does not ship Claude Code skills. (Version 2.1 added optional skills support via a standards_as_claude_code_skills config flag — that feature was retired in v3.)

Subagents

(none)

Version 3 explicitly retired subagent orchestration. The CHANGELOG states: "Implementation/orchestration phases retired — frontier models handle this well on their own now."

Hooks

(none)

No Claude Code hooks (PreToolUse, PostToolUse, UserPromptSubmit, SessionStart, Stop, SubagentStop, etc.) are defined or shipped.

MCP Servers

(none)

No MCP servers are bundled or required.

Scripts / Binaries

Agent OS ships 3 shell scripts in scripts/:

Script Purpose
project-install.sh Main install script: resolves profile inheritance, copies standards + commands into a project, builds index.yml
sync-to-profile.sh Syncs project standards back to a named base profile (for reuse across projects); supports --new-profile to create new profiles
common-functions.sh Shared shell utility functions (print helpers, YAML parsing, directory creation) sourced by the other scripts

There is no compiled binary or globally-installed CLI executable. All invocations use explicit paths like ~/agent-os/scripts/project-install.sh.

05

Prompts

Agent OS — Prompts

Agent OS's prompt design lives entirely in the five command markdown files installed to .claude/commands/agent-os/. There are no CLAUDE.md templates, no system prompts inside install scripts, and no skill files. The commands are the prompts.


Prompting Technique: AskUserQuestion-first

Every command opens with the same structural rule: always use the AskUserQuestion tool when asking the user anything. This is not a suggestion — it appears as a bolded guideline in every single command file. The technique ensures the AI never proceeds speculatively; it always explicitly surfaces a structured question and waits for a response before the next step.


Verbatim Excerpt 1 — discover-standards.md (full file, condensed to first major section)

Source: commands/agent-os/discover-standards.md

# Discover Standards

Extract tribal knowledge from your codebase into concise, documented standards.

## Important Guidelines

- **Always use AskUserQuestion tool** when asking the user anything
- **Write concise standards** — Use minimal words. Standards must be scannable by AI agents without bloating context windows.
- **Offer suggestions** — Present options the user can confirm, choose between, or correct. Don't make them think harder than necessary.

## Process

### Step 1: Determine Focus Area

Check if the user specified an area when running this command. If they did, skip to Step 2.

If no area was specified:

1. Analyze the codebase structure (folders, file types, patterns)
2. Identify 3-5 major areas. Examples:
   - **Frontend areas:** UI components, styling/CSS, state management, forms, routing
   - **Backend areas:** API routes, database/models, authentication, background jobs
   - **Cross-cutting:** Error handling, validation, testing, naming conventions, file structure
3. Use AskUserQuestion to present the areas:

I've identified these areas in your codebase:

  1. API Routes (src/api/) — Request handling, response formats
  2. Database (src/models/, src/db/) — Models, queries, migrations
  3. React Components (src/components/) — UI patterns, props, state
  4. Authentication (src/auth/) — Login, sessions, permissions

Which area should we focus on for discovering standards? (Pick one, or suggest a different area)


Wait for user response before proceeding.

### Step 3: Ask Why, Then Draft Each Standard

**IMPORTANT:** For each selected standard, you MUST complete this full loop before moving to the next standard:

1. **Ask 1-2 clarifying questions** about the "why" behind the pattern. Use your AskUserQuestion tool for this.
2. **Wait for user response**
3. **Draft the standard** incorporating their answer
4. **Confirm with user** before creating the file
5. **Create the file** if approved

Example questions to ask (adapt based on the specific standard):

- "What problem does this pattern solve? Why not use the default/common approach?"
- "Are there exceptions where this pattern shouldn't be used?"
- "What's the most common mistake a developer or agent makes with this?"

**Do NOT batch all questions upfront.** Process one standard at a time through the full loop.

Technique highlighted: The "ask-why" loop is the core extraction engine. For every discovered pattern, the agent is instructed to probe the reason behind the pattern before writing it down. This produces richer standards that capture intent, not just syntax. The explicit instruction "Do NOT batch all questions upfront" forces a tight human-in-the-loop cycle rather than an autonomous bulk run.


Verbatim Excerpt 2 — inject-standards.md (scenario detection + output format)

Source: commands/agent-os/inject-standards.md

# Inject Standards

Inject relevant standards into the current context, formatted appropriately for the situation.

## Usage Modes

This command supports two modes:

### Auto-Suggest Mode (no arguments)

/inject-standards

Analyzes context and suggests relevant standards.

### Explicit Mode (with arguments)

/inject-standards api # All standards in api/ /inject-standards api/response-format # Single file /inject-standards api/response-format api/auth # Multiple files /inject-standards root # All standards in the root folder /inject-standards root/naming # Single file from root folder

Directly injects specified standards without suggestions.

### Step 1: Detect Context Scenario

**Three scenarios:**

1. **Conversation** — Regular chat, implementing code, answering questions
2. **Creating a Skill** — Building a `.claude/skills/` file
3. **Shaping/Planning** — In plan mode, building a spec, running `/shape-spec`

**Detection logic:**

- If currently in plan mode OR conversation clearly mentions "spec", "plan", "shape" → **Shaping/Planning**
- If conversation clearly mentions creating a skill, editing `.claude/skills/`, or building a reusable procedure → **Creating a Skill**
- Otherwise → **Ask to confirm** (do not assume)

**If neither skill nor plan is clearly detected**, use AskUserQuestion to confirm:

I'll inject the relevant standards. How should I format them?

  1. Conversation — Read standards into our chat (for implementation work)
  2. Skill — Output file references to include in a skill you're building
  3. Plan — Output file references to include in a plan/spec

Which scenario? (1, 2, or 3)


Always ask when uncertain — don't assume conversation by default.

#### Scenario: Creating a Skill

First, use AskUserQuestion to determine how to include the standards:

How should these standards be included in your skill?

  1. References — Add @ file paths that point to the standards (keeps skill lightweight, standards stay in sync)
  2. Copy content — Paste the full standards content into the skill (self-contained, but won't update if standards change)

Which approach? (1 or 2)


**If References (option 1):**

Be sure to include references to the following standards files in the appropriate location in the file(s) that make up this skill:

@agent-os/standards/api/response-format.md @agent-os/standards/api/error-handling.md @agent-os/standards/global/naming.md


Technique highlighted: Context-adaptive output. The same command produces three completely different outputs depending on whether the user is in a chat, building a skill, or shaping a spec. This prevents the common failure mode of injecting verbose content into a plan that should just reference files. The @file-path reference format for skill/plan scenarios is deliberately lightweight — it keeps token cost near zero while maintaining live linkage to standards.


Verbatim Excerpt 3 — shape-spec.md (plan mode gate + Task 1 convention)

Source: commands/agent-os/shape-spec.md

# Shape Spec

Gather context and structure planning for significant work. **Run this command while in plan mode.**

## Prerequisites

This command **must be run in plan mode**.

**Before proceeding, check if you are currently in plan mode.**

If NOT in plan mode, **stop immediately** and tell the user:

Shape-spec must be run in plan mode. Please enter plan mode first, then run /shape-spec again.


Do not proceed with any steps below until confirmed to be in plan mode.

### Step 6: Generate Spec Folder Name

Create a folder name using this format:

YYYY-MM-DD-HHMM-{feature-slug}/


Where:
- Date/time is current timestamp
- Feature slug is derived from the feature description (lowercase, hyphens, max 40 chars)

Example: `2026-01-15-1430-user-comment-system/`

### Step 7: Structure the Plan

Now build the plan with **Task 1 always being "Save spec documentation"**.

Present this structure to the user:

Here's the plan structure. Task 1 saves all our shaping work before implementation begins.


Task 1: Save Spec Documentation

Create agent-os/specs/{folder-name}/ with:

  • plan.md — This full plan
  • shape.md — Shaping notes (scope, decisions, context from our conversation)
  • standards.md — Relevant standards that apply to this work
  • references.md — Pointers to reference implementations studied
  • visuals/ — Any mockups or screenshots provided

Technique highlighted: Hard prerequisite guard ("stop immediately and tell the user"). The plan mode gate is not advisory — if the command detects it's not in plan mode, execution halts. This ensures spec shaping always happens inside Claude Code's plan mode context, which provides the correct execution semantics (no accidental code writes during planning). The "Task 1 always saves spec docs" convention creates a durable, inspectable artifact of every shaping session before a single line of code is written.


Verbatim Excerpt 4 — profiles/default/global/tech-stack.md (standards file format)

Source: profiles/default/global/tech-stack.md

# Tech Stack

## Frontend

- React 18 with TypeScript
- Tailwind CSS v4 for styling
- Vite for build tooling

## Backend

- Node.js with Express
- TypeScript

## Database

- PostgreSQL

## Other

Technique highlighted: The standard file format is intentionally minimal — no prose, no rationale, no caveats. The design philosophy explicitly states "every word costs tokens" and standards should be "scannable by AI agents without bloating context windows." The default tech-stack standard is a 14-line template, not a 200-line architecture document.


Prompting Anti-Pattern: Explicit "Bad" Example in discover-standards.md

The command file includes a side-by-side "Good vs. Bad" example that doubles as prompt engineering guidance for standards authors:

## Writing Concise Standards

**Good:**
```markdown
# Error Responses

Use error codes: `AUTH_001`, `DB_001`, `VAL_001`

\`\`\`json
{ "success": false, "error": { "code": "AUTH_001", "message": "..." } }
\`\`\`

- Always include both code and message
- Log full error server-side, return safe message to client

Bad:

# Error Handling Guidelines

When an error occurs in our application, we have established a consistent pattern for how errors should be formatted and returned to the client. This helps maintain consistency across our API and makes it easier for frontend developers to handle errors appropriately...
[continues for 3 more paragraphs]

This embedded anti-pattern guidance enforces the token-economy philosophy at the content level, not just the architecture level.

---

## Summary of Prompting Techniques

| Technique | Where Used | Effect |
|---|---|---|
| `AskUserQuestion` mandate | All 5 commands | Forces human-in-the-loop; no autonomous bulk actions |
| Ask-why before drafting | `discover-standards` | Captures intent and rationale, not just syntax |
| Context scenario detection | `inject-standards` | Adaptive output format (inline vs. reference vs. copy) |
| Hard prerequisite guard | `shape-spec` | Prevents plan-mode features from running outside plan mode |
| Task 1 = save spec docs | `shape-spec` | Guarantees spec durability before implementation begins |
| Token-economy writing rules | `discover-standards` | Teaches users to write AI-friendly standards |
| Timestamped folder naming | `shape-spec` | Makes specs chronologically discoverable across sessions |
09

Uniqueness

Agent OS — Uniqueness and Positioning

What Agent OS Does That No Other Seed Framework Does

Standards discovery from live code. The /discover-standards command's core innovation is the "ask-why" extraction loop: the agent analyzes existing code, surfaces candidate patterns, then explicitly asks the developer why each pattern exists before writing it down. This produces standards that encode intent, not just syntax — a standard that says "cursor-based pagination because the dataset is unbounded" is more useful to a future AI agent than one that just says "use cursor-based pagination."

The index.yml as a standards catalog. Rather than injecting all standards every session (token-expensive) or requiring the user to remember which standards exist (cognitive burden), Agent OS maintains a YAML index of one-line descriptions. The /inject-standards command reads only the catalog to decide which full standards are relevant, then loads only those. This is a small-but-durable architecture decision not seen in other frameworks.

Context-adaptive injection format. /inject-standards detects whether it is being run in a chat, a plan, or a skill-building context and outputs different formats accordingly — inline content, @file references, or copied content. No other framework in this analysis set distinguishes injection context.

What Agent OS Explicitly Drops (vs. Other Frameworks)

Agent OS v3 deliberately removed:

  • Subagent orchestration — No orchestrate-tasks command, no agent delegation. Explicitly retired in v3.
  • Task list generation — No /create-tasks command. Delegated to the AI tool's native task tracking.
  • Implementation phase commands — No /implement-tasks. The developer uses Claude Code directly.
  • Write-spec as a separate phase — Merged into plan mode via /shape-spec.
  • Claude Code Skills conversion — v2.1 had a standards_as_claude_code_skills toggle; v3 removed it.
  • Multi-agent mode configuration — The use_claude_code_subagents config option was retired.

The CHANGELOG is candid about why: "AI coding tools have evolved significantly... Claude Code's plan mode, extended thinking, and improved models now handle much of the scaffolding that earlier versions provided."

One-Sentence Positioning

Agent OS is the framework that treats standards documentation — not orchestration, not scaffolding, not workflows — as the single highest-leverage AI coding investment, and builds exactly enough tooling around that belief.

Comparison to Similar Frameworks

Dimension Agent OS Typical orchestration framework
Core artifact agent-os/standards/*.md files Tasks, specs, agents
Token strategy Inject only what's relevant via index.yml Load CLAUDE.md always
Human loop Mandatory AskUserQuestion at every step Often autonomous
Complexity ~5 commands + 3 scripts Dozens of components
Philosophy Standards first; defer implementation to AI tools Control the full workflow

Failure Modes and Criticisms

  1. Standards quality depends entirely on the developer. The framework extracts and encodes what the developer tells it. If the developer writes vague standards ("use good naming"), the AI gets vague guidance. The framework has no validation mechanism.

  2. No enforcement. Standards are injected on demand; they are not enforced. A developer who forgets to run /inject-standards before starting work gets no standards in context. There is no hook, no CLAUDE.md always-load, no pre-commit check.

  3. Cursor support is second-class. The AskUserQuestion tool is Claude Code-specific. Cursor users get the standards storage but not the interactive command workflows.

  4. Zero issues on GitHub does not mean zero bugs. The CHANGELOG "Unreleased" section lists multiple bug fixes (set -e failures, macOS tac incompatibility) that affected production installs before being fixed. These were found by community members, not caught by tests.

  5. v3's simplification is a bet. By removing orchestration phases, Agent OS bets that Claude Code and frontier models will keep improving. If those tools regress or fragment, v3 will have less to offer than v2.

Inspired By

(none stated in repository)

Competitors Named in README

(none named directly; Antigravity is listed as a compatible tool, not a competitor)

04

Workflow

Agent OS — Workflow

Workflow Phases

Agent OS defines two distinct lifecycle tracks: project setup (run once per project) and feature development (run per feature). The v3 redesign removed a dedicated "implement" phase in favor of delegating to the AI tool's native plan mode.

Track 1: Project Setup

Phase Command Artifact Produced
1. Plan Product /plan-product agent-os/product/mission.md, roadmap.md, tech-stack.md
2. Discover Standards /discover-standards agent-os/standards/<folder>/<standard>.md files
3. Index Standards /index-standards agent-os/standards/index.yml (auto-updated after each discover run)

Track 2: Feature Development (per-feature loop)

Phase Command / Action Artifact Produced
1. Inject Standards /inject-standards Standards loaded into active context
2. Shape Spec /shape-spec (in plan mode) agent-os/specs/YYYY-MM-DD-HHMM-<slug>/plan.md, shape.md, standards.md, references.md, visuals/
3. Implement AI tool's native plan/task mode Code committed to repo

The spec folder naming convention is: agent-os/specs/YYYY-MM-DD-HHMM-feature-slug/ — e.g., 2026-01-15-1430-user-comment-system/.

Human Approval Gates

Agent OS is heavily interactive by design. Every command uses the AskUserQuestion tool for explicit human confirmation before taking actions. Key gates:

  1. /discover-standards Step 1 — User chooses which codebase area to analyze.
  2. /discover-standards Step 2 — User selects which discovered patterns to document (shown as a numbered list with opt-in/opt-out).
  3. /discover-standards Step 3 — For each selected standard, the user answers 1-2 clarifying "why" questions; the draft is confirmed before the file is written.
  4. /discover-standards Step 4 — User confirms or edits the index description for each new standard.
  5. /inject-standards — User confirms which suggested standards to inject; confirms injection format (inline content vs. @file references) for plan and skill scenarios.
  6. /index-standards — User approves one-line descriptions for newly indexed standards.
  7. /plan-product — Sequential Q&A covering problem, users, differentiator, MVP, roadmap, and tech stack; each answer is confirmed before writing files.
  8. /shape-spec — User confirms scope, visuals, code references, product alignment, and relevant standards; entire plan structure is confirmed before execution begins.
  9. Install script — If an existing agent-os/standards/ folder is detected, the script prompts before overwriting.

TDD Enforcement

No. Agent OS does not enforce or mention test-driven development. There is no test phase, no reference to test files, and no testing/ standard template. The framework is agnostic about testing methodology.

Multi-Agent Execution

No (in v3). The CHANGELOG explicitly states that subagent orchestration was retired in v3:

"Implementation/orchestration phases retired — frontier models handle this well on their own now."

Version 2.1 had a use_claude_code_subagents config toggle; v3 removed the orchestration layer entirely and defers to the AI tool's built-in task management.

Git Worktrees / Isolated Workspaces

No. Agent OS has no concept of git worktrees or isolated workspaces. It is workspace-agnostic.

Spec Format

Markdown. All spec artifacts (plan.md, shape.md, standards.md, references.md) are plain markdown. The standards index (index.yml) is YAML but it is a discovery index, not a spec format.

Files Generated Per Feature

File Location Contents
plan.md agent-os/specs/<folder>/ The full implementation plan (Task 1 always saves spec docs first)
shape.md agent-os/specs/<folder>/ Shaping decisions: scope, constraints, product alignment, standards applied
standards.md agent-os/specs/<folder>/ Full content of the relevant standards for this feature
references.md agent-os/specs/<folder>/ Pointers to similar/reference code in the codebase
visuals/ agent-os/specs/<folder>/visuals/ Mockups and screenshots (optional; created if visuals provided)

The spec folder slug is derived from the feature description: lowercase, hyphens, max 40 characters.

06

Memory Context

Agent OS — Memory and Context

Memory Model

file-based

Agent OS persists all state as plain markdown and YAML files in the project repository. There is no database, no vector store, no external memory service, and no MCP bridge.

Persistence Scope

project

All artifacts live inside the project directory under agent-os/. They are checked into (or at least co-located with) the project repo. There is no global memory across different projects.

Files That Constitute Memory

File / Path Role
agent-os/standards/index.yml Discovery index: maps every standard file to a one-line description; the AI reads this to decide which standards are relevant without loading all files
agent-os/standards/<folder>/<name>.md The actual standards — codebase conventions, patterns, constraints
agent-os/product/mission.md Product mission, target users, differentiator
agent-os/product/roadmap.md MVP and post-launch feature list
agent-os/product/tech-stack.md Technology choices
agent-os/specs/YYYY-MM-DD-HHMM-<slug>/plan.md Timestamped feature plan
agent-os/specs/YYYY-MM-DD-HHMM-<slug>/shape.md Shaping decisions and context from the planning conversation
agent-os/specs/YYYY-MM-DD-HHMM-<slug>/standards.md Copy of the standards that applied to this feature (snapshot at shaping time)
agent-os/specs/YYYY-MM-DD-HHMM-<slug>/references.md Pointers to similar code studied during shaping

Context Compaction Strategy

Agent OS does not address /compact or context loss directly. Its architecture is the compaction strategy: standards are kept deliberately minimal (token-economy writing rules are enforced), and they are only loaded on demand via /inject-standards rather than always present. The index.yml acts as a lightweight catalog — the AI reads only the 1-line descriptions to pick which full standards to load, avoiding bulk-loading all standards at once.

Cross-Session Handoffs

Yes — by design, through the timestamped spec folder convention. When a new session begins on a feature, the developer runs /shape-spec which finds the existing agent-os/product/ context and the AI can be directed to read the relevant spec folder. The shape.md file captures:

  • Scope decisions
  • Key constraints
  • Visuals references
  • Standards that applied

This gives the next session (or a new agent subagent) enough context to resume without re-interviewing the developer. The spec folder's timestamp makes it easy to find the most recent shaping session for a feature.

Explicit References to Memory Concepts

  • The system uses the phrase "tribal knowledge" — the target artifact of /discover-standards is to convert tribal knowledge into documented standards.
  • The shape.md file is explicitly described as capturing "decisions made during shaping" and "context" — functioning as a session handoff document.
  • Standards files are described as living alongside the codebase long-term; they are intended to be committed to the repository and persist across team members and sessions.

Memory Not Present

  • No "memory bank" named construct.
  • No cross-project memory (each project's agent-os/ is independent).
  • No vector similarity search for standards retrieval (text matching via index.yml descriptions instead).
  • No conversation history storage.
07

Target Tools

Agent OS — Target Tools

Officially Supported Tools

Claude Code

Evidence: The README states "Works alongside Claude Code, Cursor, Antigravity, and other AI tools." The install script explicitly places commands in .claude/commands/agent-os/, which is the Claude Code slash-command discovery path. The CHANGELOG and command files extensively reference Claude Code features:

  • Plan mode (required for /shape-spec)
  • AskUserQuestion tool (used in every command)
  • Claude Code Skills (.claude/skills/) referenced in /inject-standards output
  • Previous v2.1 had a use_claude_code_subagents config option

Install path: Commands land at <project>/.claude/commands/agent-os/*.md (5 files).

Compatibility: Full native support. All commands are designed around Claude Code's plan mode and tool execution model.


Cursor

Evidence: The README explicitly lists Cursor: "Works alongside Claude Code, Cursor, Antigravity, and other AI tools."

Install path: The project-install script does not have a Cursor-specific code path. Standards land at <project>/agent-os/standards/ (same as Claude Code). Command files would need to be placed manually in Cursor's rules directory (.cursor/rules/ or equivalent). The README notes "Docs, installation, usage, & best practices 👉 It's all here" pointing to buildermethods.com/agent-os for Cursor-specific guidance.

Compatibility: Partially supported. Standards storage is identical. The command markdown files can be used as Cursor rules or manually injected prompts, but they won't work as slash-commands natively.

Caveat: The AskUserQuestion tool references in command files are Claude Code-specific. Cursor users would need to adapt these prompts to Cursor's equivalent interaction model.


Antigravity

Evidence: Listed in the README: "Works alongside Claude Code, Cursor, Antigravity, and other AI tools."

Install path: No Antigravity-specific installation path documented in the repository. Standards storage would be the same. Commands would require manual integration.

Compatibility: Unknown — no specific instructions found in the repository source.


"Other AI Tools"

The README uses the open phrase "other AI tools" and "Any language, any framework." The underlying mechanism (markdown standards files + YAML index) is tool-agnostic. Any AI tool that can read files and accept a system context would be compatible with the standards layer.

Summary Table

Tool Explicitly Named Commands Auto-Install Standards Compatible Notes
claude-code Yes Yes (.claude/commands/) Yes Primary supported tool; plan mode required for /shape-spec
cursor Yes No Yes Standards same; command integration manual
antigravity Yes No Yes No specific docs in repo
other Implied No Yes Any tool that reads files

Primary Tool

claude-code

All commands use Claude Code-specific tooling (AskUserQuestion, plan mode detection, .claude/commands/ path) and the CHANGELOG explicitly tracks Claude Code feature adoption as a reason for architectural changes.

08

Signals

Agent OS — Signals

GitHub Metrics

Metric Value Source
Stars 4,646 gh api /repos/buildermethods/agent-os (fetched 2026-05-26)
Forks 740 Same API call
Watchers (subscribers) 103 Same API call
Contributors 17 gh api /repos/buildermethods/agent-os/contributors
Last commit 2026-05-05 Latest commit on main branch
First commit 2025-07-16 Repository creation date
Open issues 0 GitHub API
Releases 12 GitHub releases page
Latest release v3.0.0 (2026-01-20) GitHub releases

Maintainer Status

Active. The project has 130 commits, 12 releases, and the most recent commit was May 5, 2026 (21 days before this analysis). The CHANGELOG has an "Unreleased" section with bug fixes being tracked. Brian Casel is identified as the primary maintainer and is actively publishing documentation and videos at buildermethods.com.

Community Activity

The repository has 0 open issues and 0 open pull requests as of the analysis date. This is consistent with a project where the maintainer reviews and merges quickly, or where external contributions are handled via discussions. There is a Discussions tab mentioned in the repository navigation.

Reddit / HN Sentiment

Unknown. No Reddit or HN quotes for Agent OS were found in the referenced _index/wave-2b-reddit-discovery.md analysis seed. The high star count (4,646) and fork count (740) relative to the project's age (launched July 2025, ~10 months old) indicates strong community interest. The commit message "Brian Casel (via Claude)" on recent commits is a notable signal: the maintainer uses Claude directly to contribute to the project.

Velocity Observations

  • ~13 commits/month average over 10 months of existence.
  • v3.0 was a major breaking change that stripped out ~half the features (implementation/orchestration phases), which is an unusual but deliberate product decision.
  • The scripts/ directory language is listed as 100% Shell on GitHub — entirely bash, no higher-level language dependencies.

Related frameworks

same archetype · same primary tool · same memory type

BMAD-METHOD ★ 48k

Provides a full agile delivery lifecycle with named expert-persona AI collaborators that elicit the human's best thinking rather…

Claude Conductor ★ 367

Gives Claude Code a persistent, cross-linked, auto-analyzed documentation system so it retains codebase context across sessions.

Spec-Driver (Greenfield Spec-Driven Development) ★ 25

Prevents spec rot in AI-assisted development by making implementation changes flow back into evergreen, authoritative specs via…

Anthropic Knowledge Work Plugins ★ 16k

Role-specialized plugin bundles with live MCP connectors that turn Claude into a domain expert for enterprise knowledge workers.

Codex Integration for Claude Code (skill-codex) ★ 1.3k

Single Claude Code skill that handles Codex CLI invocation correctly (stdin blocking, thinking token suppression, session resume)…

mini-coding-agent ★ 882

A single-file zero-dependency Python coding agent that demonstrates the six core components of coding agents for educational…