Skip to content

🔌 Adapters

SpecMem adapters parse specifications from various AI coding agent frameworks.

Supported Frameworks

Spec-Driven Development Frameworks (Priority)

Framework Adapter File Patterns
Kiro kiro .kiro/specs/**/*.md
OpenSpec openspec openspec/specs/**, openspec/changes/**, openspec/config.yaml
SpecKit speckit .speckit/**/*.yaml
Tessl tessl .tessl/**/*.md

Commercial Coding Agents

Framework Adapter File Patterns
AGENTS.md (AAIF) agents.md **/AGENTS.md, **/AGENT.md
Claude Code claude Claude.md, CLAUDE.md
Cursor cursor .cursorrules, cursor.rules, .cursor/rules/*.mdc

Codex, Factory, Warp, Cursor, OpenCode, Amp, and Aider consume AGENTS.md. SpecMem indexes those files through the AGENTS.md adapter rather than invented vendor trees (.codex/**/*.md, .factory/**/*.yaml, .warp/**). Gemini CLI's GEMINI.md is found by the guidelines scanner, not a separate SpecAdapter.

Spec-Driven Development Frameworks

These frameworks prioritize formal specifications and structured development workflows.

Kiro Adapter

Parses Kiro's spec-driven development format. Kiro is the recommended framework for teams adopting Spec-Driven Development.

Structure

.kiro/
└── specs/
    └── feature-name/
        ├── requirements.md
        ├── design.md
        └── tasks.md

Example

# Requirements Document

## Introduction
User authentication system for the application.

## Requirements

### Requirement 1
**User Story:** As a user, I want to log in securely.

#### Acceptance Criteria
1. WHEN a user enters valid credentials THEN the system SHALL authenticate
2. WHEN a user enters invalid credentials THEN the system SHALL reject

Configuration

[adapters]
kiro = true

[adapters.kiro]
spec_dir = ".kiro/specs"

OpenSpec Adapter

Parses Fission-AI OpenSpec on-disk layout. This adapter is stable (not experimental): the official directory structure and artifact filenames are documented and clear.

Structure

openspec/
├── specs/                 # Source of truth (often <domain>/spec.md)
│   └── <domain>/
│       └── spec.md
├── changes/               # Active (and archived) change folders
│   └── <change-name>/
│       ├── proposal.md
│       ├── design.md
│       ├── tasks.md
│       └── specs/         # Delta specs
│           └── <domain>/
│               └── spec.md
└── config.yaml            # Optional planning context / rules

Type mapping

Path / filename SpecType
specs/**/spec.md, change delta specs/** requirement
design.md design
tasks.md task
proposal.md, other markdown knowledge
config.yaml (context, rules, …) knowledge

Markdown is split on # / ## / ### headings (same approach as AGENTS.md / Cursor). Blocks are tagged openspec plus path-derived slugs. Files under changes/archive/ are loaded as legacy. Junk directories and openspec/schemas/ are skipped.

Configuration

[adapters]
# Auto-discovered; no extra config required
openspec = true

SpecKit Adapter

Parses SpecKit's YAML-based specifications. SpecKit provides structured, machine-readable specs.

Structure

.speckit/
├── features/
│   └── auth.yaml
├── constraints/
│   └── security.yaml
└── config.yaml

Example

# .speckit/features/auth.yaml
feature: user-authentication
version: 1.0.0

requirements:
  - id: AUTH-001
    title: User Login
    description: Users can log in with email and password
    priority: critical
    acceptance_criteria:
      - Valid credentials grant access
      - Invalid credentials show error
      - Account lockout after 5 failures

  - id: AUTH-002
    title: Password Reset
    description: Users can reset forgotten passwords
    priority: high

Configuration

[adapters]
speckit = true

[adapters.speckit]
spec_dir = ".speckit"

Tessl Adapter

Parses Tessl's markdown specifications. Tessl focuses on test-driven specifications.

Structure

.tessl/
├── specs/
│   └── feature.md
└── config.toml

Configuration

[adapters]
tessl = true

[adapters.tessl]
spec_dir = ".tessl/specs"

Commercial Coding Agents

These adapters support popular commercial AI coding assistants.

Claude Code Adapter

Parses Claude Code's markdown-based specifications.

Files

  • Claude.md or CLAUDE.md - Project context and rules

Example

# Project Context

## Overview
E-commerce platform built with Python and FastAPI.

## Architecture
- Backend: FastAPI with SQLAlchemy
- Database: PostgreSQL
- Cache: Redis
- Queue: Celery

## Coding Standards
- Type hints required for all functions
- Docstrings in Google format
- 100% test coverage for business logic

## Security Requirements
- All endpoints require authentication
- Input validation on all user data
- SQL injection prevention via ORM

Configuration

[adapters]
claude = true

[adapters.claude]
files = ["Claude.md", "CLAUDE.md"]

Cursor Adapter

Parses Cursor rules files. This adapter is experimental.

Files

  • .cursor/rules/*.mdc — current Cursor project rules (YAML frontmatter + markdown)
  • .cursorrules — legacy single-file rules
  • cursor.rules — legacy alternate filename

Example .cursor/rules/*.mdc

---
description: Frontend component standards
globs: src/components/**/*.tsx,src/**/*.ts
alwaysApply: false
---

# Components

Use the design system.

## Validation

Validate API inputs.

Frontmatter fields mapped onto SpecBlocks (without new SpecIR fields):

  • alwaysApply: true → tag always-apply
  • description → tag described and text prefix
  • globs → truncated globs:... tag and a Globs: text line

Example .cursorrules

# Project Rules

## Code Style
- Use 2-space indentation
- Prefer const over let
- Use async/await over promises

## Architecture
- Follow clean architecture principles
- Keep components under 200 lines
- Extract reusable logic to hooks

Configuration

[adapters]
cursor = true

AGENTS.md Adapter

Parses the AAIF / Linux Foundation AGENTS.md standard. This is a stable adapter (not experimental). Codex, Factory, Warp, Cursor, OpenCode, Amp, and Aider all consume these files.

Files

  • AGENTS.md or Agents.md — official filename (repo root or nested)
  • AGENT.md or Agent.md — aliases also accepted by the guidelines scanner

Closest-file-wins: a nested AGENTS.md applies to that subtree.

Example

# Project Rules

Keep pull requests focused and scoped to one change.

## Testing

Run unit tests before opening a PR.

## Style

Prefer typed Python.

Heading sections become knowledge SpecBlocks tagged agents / agents.md plus a slug of the heading. A file with no headings becomes one pinned block.

Configuration

[adapters]
# Auto-discovered; no extra config required

There are no first-class Codex, Factory, Warp, or Gemini SpecAdapters. Those tools are covered by AGENTS.md (and, for Gemini CLI, the guidelines scanner's GEMINI.md support). Do not invent .codex/, .factory/, or .warp/ trees.

Using Multiple Adapters

SpecMem can use multiple adapters simultaneously:

[adapters]
# Spec-Driven Development (recommended)
kiro = true
openspec = true
speckit = true
tessl = true

# Commercial agents
claude = true
cursor = true

When scanning, SpecMem merges specs from all enabled adapters:

specmem scan

🔍 Scanning for specifications...
✅ [kiro] Found 12 specs in .kiro/specs/
✅ [speckit] Found 5 specs in .speckit/
✅ [tessl] Found 3 specs in .tessl/
✅ [claude] Found 1 spec in Claude.md
✅ [cursor] Found 3 specs in cursor.json
📊 Total: 24 specifications indexed

Adapter Priority

When specs conflict, priority is determined by:

  1. Explicit priority in the spec
  2. Framework order (configurable)
  3. Timestamp (newer wins)
[adapters]
# Framework priority (first = highest)
# Spec-Driven Development frameworks take precedence
priority = ["kiro", "openspec", "speckit", "tessl", "agents.md", "claude", "cursor"]

Custom Adapters

See Writing Adapters for creating custom adapters.