Agent-ready documentation for Taskflow, a header-only C++20 parallel task programming library. Built on the AGENTS.md open standard.
AGENTS.md is an open format for guiding AI coding agents — think of it as a README for AI. Instead of cluttering your human-facing README with build commands, test steps, and code conventions that only agents need, you put them in an AGENTS.md file that coding agents automatically discover and follow.
The format is stewarded by the Agentic AI Foundation under the Linux Foundation, with backing from OpenAI, Google, Cursor, Anthropic, and others. Over 60,000+ open-source projects already use it.
| File | Description |
|---|---|
AGENTS.md |
Root agent instructions — core API, patterns, cost model, common mistakes |
docs/taskflow-patterns.md |
Code conversion recipes, advanced patterns, large-scale DAGs |
docs/taskflow-performance.md |
Profiling (DigestObserver), per-task cost model, tuning, NUMA pinning |
docs/taskflow-internals.md |
Contributing to Taskflow source code |
digest_observer.hpp |
Standalone diagnostic profiler — single-header, ~2 KB structured digest in O(1) memory |
INTEGRATION.md |
Integration guide for digest_observer.hpp (standalone use, in-tree integration, multi-phase apps) |
field-notes/opentimer-granularity-experiments.md |
OpenTimer case study using the digest profiler — profile→hypothesize→experiment→re-profile; NUMA + SMT findings |
field-notes/taskbench-porting.md |
Porting a Taskflow runtime into the task-bench benchmark framework |
When an agent works on Taskflow code, it reads AGENTS.md to understand the library's API, patterns, and pitfalls — then writes better code on the first try.
Copy AGENTS.md and the docs/ folder into the root of any project that uses Taskflow. Your agent will pick them up automatically.
Claude Code automatically reads AGENTS.md from your project root. Just drop the files in and start coding:
# Copy into your project
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
# Start Claude Code in your project
cd /path/to/your-project
claudeClaude Code will automatically discover and follow the instructions in AGENTS.md. You can also reference the docs explicitly:
> Help me parallelize this loop using Taskflow
> Convert this OpenMP code to use Taskflow's reduce_by_index
Codex automatically reads AGENTS.md when present in the repository root. No extra configuration needed:
# Copy into your project
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
# Codex will pick up AGENTS.md automatically
codexGoogle Jules reads AGENTS.md from the project root automatically. Place the files and Jules will follow the conventions when working on your Taskflow code:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/Cursor natively supports AGENTS.md. Drop the file into your project root and Cursor's AI features (Cmd+K, Chat, Tab completions) will use it as context:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
# Open in Cursor
cursor /path/to/your-projectWindsurf reads AGENTS.md automatically from the project root. No configuration required:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/OpenCode discovers AGENTS.md from the working directory. Place the files and start a session:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
cd /path/to/your-project
opencodeGitHub Copilot reads AGENTS.md when it's present in your workspace root. Works in both VS Code and JetBrains IDEs:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
# Open in VS Code or JetBrains — Copilot will use the agent docs
code /path/to/your-projectAider needs a one-line config to read AGENTS.md. Add to .aider.conf.yml in your project root:
read: AGENTS.mdThen run Aider as usual:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
cd /path/to/your-project
aiderZed's AI assistant automatically picks up AGENTS.md from the project root. Just open your project in Zed:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
zed /path/to/your-projectConfigure Gemini CLI to read AGENTS.md by adding a .gemini/settings.json file:
{
"contextFileName": "AGENTS.md"
}Then:
cp AGENTS.md /path/to/your-project/
cp -r docs/ /path/to/your-project/docs/
cd /path/to/your-project
geminiDevin automatically reads AGENTS.md when it's present in the repository. No setup required — just include the files in your repo.
Amp automatically discovers and reads AGENTS.md from the project root. No configuration needed.
You don't need a local coding agent to benefit from these docs. You can use them directly in web-based AI tools:
- Copy the contents of
AGENTS.mdinto your conversation - Ask the AI to help you write Taskflow code using those conventions
- The AI will follow the patterns, avoid common mistakes, and use the right APIs
Example prompt:
Here are my project's Taskflow conventions: [paste AGENTS.md contents]
Help me convert this sequential loop to use Taskflow's for_each_by_index with a DynamicPartitioner.
If AGENTS.md is committed to your repository, GitHub Copilot Chat on github.com will use it as context when you ask questions about your code.
Cloud development environments with AI features will automatically pick up AGENTS.md from the workspace, just like local editors.
AGENTS.md supports hierarchical placement. The file closest to the code being edited takes precedence. For monorepos with multiple Taskflow-based subprojects, you can place tailored AGENTS.md files in each subdirectory:
my-monorepo/
AGENTS.md # shared conventions
services/
compute-engine/
AGENTS.md # Taskflow-specific guidance (this repo's files)
web-api/
AGENTS.md # different conventions for the web service
- AGENTS.md specification — the open standard
- Taskflow documentation — official Taskflow docs
- Taskflow GitHub — source code