GitHubSecurityLab

seclab-taskflow-agent: Build YAML-Driven AI Agent Workflows

GitHub Security Lab’s Taskflow Agent is an MCP-enabled, multi-agent framework for defining workflows in YAML. Built on OpenAI’s Agents SDK, it targets security research and vulnerability triage without requiring users to write workflow code.

ActivePythonMITAI AgentsSecurityMCP
View on GitHub Compare
Stars
285
Forks
42
Stars, last 30 days
+10
Last commit
today
Contributors
12

Our take

Activeregular commits and releases

Health score

51/100

How it is scored
Activity
65
Community
54
Issues
21
Pull requests
64

+10 stars in the last 30 days

  • Releases more often than 79% of the projects we track

This is an actively maintained, specialized framework for teams willing to configure agent workflows and MCP tools, with recent releases and contributions from multiple people.

Good fit if

  • Your security research or triage work benefits from repeatable, YAML-defined multi-agent workflows.
  • You need MCP tool integration, task-level recovery, or offline workflow validation.
  • You can provide Python 3.10 or later and the model-service credentials your chosen setup requires.

Look elsewhere if

  • You need a general-purpose agent product that does not require YAML configuration or MCP setup.
  • You need consistent features across SDK backends, since supported capabilities differ.
  • You cannot use the MIT-licensed project or provide the required model API credentials.
All health signals
Last commit2026-11-02 (today)
Commits, last 90 days19
Releases, last 12 months28 (latest v0.5.1, 2026-10-16)
Contributors12 (top contributor: 49% of commits)
Issues closed, last 90 days0
Pull requests merged, last 90 days9 (typically merged in 0 days)
Project age14 months

Checked on 2026-11-03 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

288

Forks

42

Change since Nov 21, 2026: 0 stars·0 forks

Overview

GitHub Security Lab’s Taskflow Agent is an MCP-enabled, multi-agent framework for declarative, YAML-driven workflows. Built on OpenAI’s Agents SDK, it lets teams compose agents, tasks, and MCP tools without writing workflow code, addressing the need to repeat and adapt agent-driven security work as prompts and models change.

The project is aimed particularly at security researchers and teams doing code auditing or vulnerability triage. Its YAML grammar supports task sequencing, agent handoffs, templates, model configuration, checkpointing, and offline linting. It also offers Copilot and Anthropic SDK backends, with capabilities that vary by backend.

Key Features

  • Define taskflows, agent personalities, reusable prompts, toolboxes, and model configurations in versioned YAML files.
  • Connect agents to MCP tools over supported transports, including stdio and HTTP-based options.
  • Use agent handoffs for multi-step work, or fan tasks out across models and compare their captured responses.
  • Configure OpenAI Agents, GitHub Copilot, or Anthropic SDK backends, with explicit capability checks for unsupported fields.
  • Retry failed tasks, checkpoint runs, and resume from the last successful task.
  • Lint taskflows and their references offline, with an option to treat unknown fields as errors.
  • Generate run manifests with task status, model, timing, and output information.

Use Cases

  • Security researchers can use CodeQL MCP tools and targeted prompts to explore code and review potential vulnerabilities.
  • Vulnerability-triage teams can define repeatable workflows for assessing incoming findings or scanning alerts.
  • Model evaluators can send the same question to multiple models and use a follow-up task to compare their answers.
  • Teams can prototype agent workflows in YAML and validate their configuration in CI before making model calls.

What you need

Detected in the repository

  • Python >=3.10 (from pyproject.toml)
  • A test suite and automated checks on GitHub Actions

License in plain words

MITpermissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the copyright and license notice
  • Share your changes: no

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

From a checkout, build and run with Hatch:

pip install hatch
hatch build
hatch run main -t examples.taskflows.example

Set AI_API_TOKEN for a GitHub Copilot-entitled account, then configure any additional credentials and MCP services your taskflow uses. See the README for configuration, Docker usage, and more examples.

Alternatives

  • autogen: AutoGen is a general-purpose Python framework for multi-agent orchestration, while Taskflow defines security research and triage workflows in YAML.
  • ruflo: Ruflo coordinates Claude Code and Codex agents for coding workflows, while Taskflow targets security research and vulnerability triage with YAML-defined workflows.
ProjectLanguageLicenseStarsStatus
seclab-taskflow-agentPythonMIT285Active
autogenPythonCC-BY-4.061.3kQuiet
rufloTypeScriptMIT73.8kActive

Considerations

The project describes itself as experimental, so teams should test workflows and backend behavior against their own tasks before relying on them. It is actively developed, with regular commits and releases, and contributions are not concentrated in a single person; however, no issues were closed in the measured recent period. Backend differences matter: for example, the Copilot SDK does not honor the YAML api_type setting and does not support multi-personality handoffs. Workflows also depend on external model credentials and whichever MCP servers they configure. The README warns that the Docker image is a deployment convenience, not a security boundary, and recommends controlling which environment variables MCP subprocesses inherit.

Found this useful?

Share it with someone who would like seclab-taskflow-agent.

Comparisons