manus-agent

A powerful, extensible framework for building AI agents with comprehensive tool support, multi-agent orchestration, and advanced vulnerability intelligence.

Built on Strands SDK and integrated with browser-use.

Installation

# Basic installation
pip install manus-agent

# With browser automation support
pip install manus-agent[browser]
playwright install chromium

# Full installation with all optional features
pip install manus-agent[browser,search,visualization]

Quick Start

# 1. Set up credentials
manus-agent init

# 2. Verify your environment
manus-agent doctor

# 3. Run a task
manus-agent "Write a Python script that fetches the current Bitcoin price"

# Or start the interactive REPL
manus-agent

CLI Reference

`manus-agent [task]` — Run a task

# Single-shot (prints result, then exits)
manus-agent "Create a factorial function in Python"

# Use a specific agent type
manus-agent --agent browser "Find the top 5 trending GitHub repos today"

# Force multi-agent orchestration
manus-agent --mode multi "Research quantum computing and create a presentation"

# JSON output for piping
manus-agent --format json "List the first 10 prime numbers" | jq .result

# Stream output tokens in real time
manus-agent --stream "Write a short story about a robot"

# Interactive REPL
manus-agent
manus-agent --mode multi

Flag	Default	Description
`--mode {auto,single,multi}`	`auto`	Execution mode; `auto` detects task complexity
`--agent {manus,browser,data,mcp}`	`manus`	Agent type for single-agent mode
`--show-plan`	off	Print the multi-agent plan before running
`--output FILE`	—	Write result to FILE (single-shot only)
`--format {text,json}`	`text`	Output format
`--stream`	off	Stream output tokens in real time
`--no-history`	off	Skip recording this run in the history log
`--config FILE`	—	Override config file path
`--version`	—	Print version and exit

`manus-agent init` — Configure credentials

manus-agent init                        # write to ~/.manus-agent/config.toml
manus-agent init --output ./my.toml    # write to a custom path
manus-agent init --force                # overwrite without prompting

`manus-agent doctor` — Diagnose your environment

manus-agent doctor
manus-agent doctor --config ./custom.toml

Checks Python packages, config file validity, and API key accessibility.

`manus-agent history` — Browse past runs

manus-agent history                        # last 20 runs
manus-agent history --limit 50            # last 50 runs
manus-agent history --grep "bitcoin"      # filter by task text
manus-agent history --format json | jq .  # all history as JSON
manus-agent history --clear               # delete all history

History is stored at ~/.manus-agent/history.jsonl.

`manus-agent analyze <CVE-ID>` — Vulnerability intelligence

manus-agent analyze CVE-2025-6554
manus-agent analyze CVE-2024-3094 --verify
manus-agent analyze CVE-2025-6554 --output json
manus-agent analyze CVE-2025-6554 --output lark

Runs an 8-step intelligence pipeline — see Security & Vulnerability Intelligence for full details.

Flag	Default	Description
`--verify`	off	Run exploit in a Docker sandbox to confirm exploitability
`--output {text,json,lark}`	`text`	Report format
`--config FILE`	—	Override config

`manus-agent remediate <CVE-ID>` — Remediation guidance

manus-agent remediate CVE-2024-3094
manus-agent remediate CVE-2024-3094 --output json

Flag	Default	Description
`--output {text,json}`	`text`	Report format
`--config FILE`	—	Override config

`manus-agent discover` — CVE discovery

manus-agent discover
manus-agent discover --since 2025-06-01 --min-epss 0.7
manus-agent discover --dry-run
manus-agent discover --output json

Flag	Default	Description
`--since YYYY-MM-DD`	4 weeks ago	Start date for the discovery window
`--min-epss SCORE`	`0.5`	Minimum EPSS score (0.0–1.0)
`--output {text,json}`	`text`	Report format
`--dry-run`	off	Discover CVEs but do not submit them
`--config FILE`	—	Override config

`manus-agent epss-trend <CVE-ID>` — EPSS score history

manus-agent epss-trend CVE-2024-3094
manus-agent epss-trend CVE-2024-3094 --days 90
manus-agent epss-trend CVE-2024-3094 --output json | jq .analysis.spike_detected

Fetches daily EPSS scores from the FIRST.org API and detects exploitation spikes (≥ 0.10 jump in a 7-day window).

Flag	Default	Description
`--days N`	`30`	Days of history (max 365)
`--output {text,json}`	`text`	Output format

`manus-agent patch-diff <CVE-ID>` — Patch diff summariser

manus-agent patch-diff CVE-2024-3094
manus-agent patch-diff CVE-2024-3094 --output json | jq .commit_summaries

Finds the fixing commit(s) via GHSA + NVD, fetches the raw unified diff, and produces a structured summary: files/functions changed, bug class (14 categories), reproduction condition hints, and commit URL.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent variants <CVE-ID>` — Variant analysis

manus-agent variants CVE-2024-3094
manus-agent variants CVE-2024-3094 --output json

Flag	Default	Description
`--output {text,json}`	`text`	Report format

`manus-agent compare <CVE-A> <CVE-B>` — Side-by-side comparison

manus-agent compare CVE-2024-3094 CVE-2021-44228
manus-agent compare CVE-2024-3094 CVE-2021-44228 --output json | jq .higher_priority

Fetches NVD, EPSS, and CISA KEV data for both CVEs in parallel and produces a side-by-side comparison across CVSS, EPSS, KEV membership, CWE, and exploitability factors. Outputs a prioritisation recommendation with confidence level (strong / moderate / weak).

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent exploit-complexity <CVE-ID>` — Exploit complexity scorer

manus-agent exploit-complexity CVE-2024-3094
manus-agent exploit-complexity CVE-2024-3094 --output json | jq .attacker_friendly

Scores practical attacker effort on a 1–5 scale across five dimensions:

Dimension	What it measures
Lines of code	How much exploit code must be written/adapted
Authentication	Credentials or privilege level required
Network hops	How many services the exploit must reach
OS/platform deps	Platform-specific syscalls, structs, gadgets
Exploit chain length	Number of discrete attack stages

Outputs a complexity_score (1–5), a label (trivial / low / moderate / high / very_high), and an attacker_friendly boolean (true when score ≤ 2.5).

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent poc-search <CVE-ID>` — Multi-source PoC aggregator

manus-agent poc-search CVE-2024-3094
manus-agent poc-search CVE-2024-3094 --sources trickest,exploitdb,github
manus-agent poc-search CVE-2024-3094 --output json | jq .exploited_in_wild

Queries five PoC sources in parallel, deduplicates results, and sorts by exploited-in-wild status and publication date:

Source	What it provides
`trickest`	trickest/cve — 250k+ CVE PoC index
`vulncheck_kev`	VulnCheck KEV — exploited-in-wild signal from 100+ intel sources (requires `VULNCHECK_API_KEY`)
`exploitdb`	Exploit-DB CSV — cached 24 h
`github`	GitHub repo search for repositories mentioning the CVE
`nvd`	NVD references filtered for GitHub / Exploit-DB / PacketStorm URLs

Flag	Default	Description
`--sources LIST`	all	Comma-separated subset: `trickest,vulncheck_kev,exploitdb,github,nvd`
`--output {text,json}`	`text`	Output format

`manus-agent blast-radius <SPEC>` — Dependency blast radius

manus-agent blast-radius requests@2.28.0
manus-agent blast-radius pypi:urllib3@1.26.5
manus-agent blast-radius npm:lodash@4.17.20
manus-agent blast-radius CVE-2021-44228
manus-agent blast-radius CVE-2021-44228 --output json | jq .summary

Estimates how broadly a vulnerability propagates downstream. Resolves affected packages from NVD CPE + OSV.dev + GHSA, then enriches each with real download/dependent stats:

Metric	Source
npm dependent packages	npm registry search API
npm weekly/monthly downloads	npm downloads API
PyPI downloads	PyPI JSON API + pypistats.org
Maven artifact metadata	Maven Central Solr search

Blast-radius labels per package:

Label	Threshold
CRITICAL	≥ 5 M weekly downloads or ≥ 50 K npm dependents
HIGH	≥ 500 K downloads or ≥ 5 K dependents
MEDIUM	≥ 50 K downloads or ≥ 500 dependents
LOW	any measurable signal
UNKNOWN	no data available

Flag	Default	Description
`--max-packages N`	`10`	Max affected packages to enrich
`--output {text,json}`	`text`	Output format

`manus-agent silent-patches <owner/repo>` — Silent patch detector

manus-agent silent-patches torvalds/linux
manus-agent silent-patches torvalds/linux --since 2025-01-01
manus-agent silent-patches torvalds/linux --output json | jq .[].classification

Scans a repository's commit history for security fixes that were never assigned a CVE. Uses two-stage heuristic scoring: commit message keywords then diff keywords. Each candidate commit is labelled with one of 14 bug classes (e.g. auth_bypass, buffer_overflow, use_after_free).

Flag	Default	Description
`--since YYYY-MM-DD`	90 days ago	Start date for commit scan
`--until YYYY-MM-DD`	today	End date
`--max-commits N`	`500`	Hard limit on commits fetched
`--fast`	off	Skip diff scoring (message keywords only)
`--output {text,json}`	`text`	Output format

`manus-agent cve-timeline <CVE-ID>` — CVE timeline

manus-agent cve-timeline CVE-2021-44228
manus-agent cve-timeline CVE-2021-44228 --output json

Reconstructs the full event timeline for a CVE: NVD publish date → EPSS history → CISA KEV add date → patch release date. Useful for understanding how quickly a vulnerability was weaponised and fixed.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent version-range <CVE-ID>` — Affected version ranges

manus-agent version-range CVE-2021-44228
manus-agent version-range CVE-2021-44228 --ecosystem pypi
manus-agent version-range CVE-2021-44228 --output json | jq .first_patched_version

Walks NVD CPE configurations and cross-references PyPI / npm / Maven to produce structured vulnerable semver ranges, a list of affected releases, and the first patched release.

Flag	Default	Description
`--ecosystem {auto,pypi,npm,maven}`	`auto`	Force a specific ecosystem
`--output {text,json}`	`text`	Output format

`manus-agent vendor-response <CVE-ID>` — Vendor response tracker

manus-agent vendor-response CVE-2024-3094
manus-agent vendor-response CVE-2024-3094 --output json | jq .classification

Queries four sources (NVD reference URL patterns, GHSA published state + patched_versions, CISA KEV required-action + due-date, repo-level GitHub security advisories) and outputs a 6-state patch-status classification:

patch_available · patch_backported · wont_fix · investigating · no_patch · unknown

Confidence is rated high / moderate / low. A VulnCheck KEV hit upgrades confidence when VULNCHECK_API_KEY is set.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent poc-freshness <CVE-ID>` — PoC freshness checker

manus-agent poc-freshness CVE-2024-3094
manus-agent poc-freshness CVE-2024-3094 --output json | jq .freshness_score

Measures how recently PoC activity occurred: last commit recency in known PoC repos, recently-starred repositories, new Exploit-DB entries. A high freshness score means attacker interest is ongoing.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent sbom-scan <bom-file>` — SBOM scanner

manus-agent sbom-scan bom.json
manus-agent sbom-scan sbom.spdx.json --output json | jq .critical_count

Parses CycloneDX or SPDX SBOMs, queries OSV.dev in batch for all components, enriches each finding with EPSS and CISA KEV status, and ranks results by KEV membership then EPSS score.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent temporal-priority <CVE-ID>` — Temporal priority scorer

manus-agent temporal-priority CVE-2024-3094
manus-agent temporal-priority CVE-2024-3094 --output json | jq .score

Produces a 0–100 urgency score combining CVSS base score, current EPSS, EPSS spike recency, CISA KEV membership, patch availability, and CVE age. Designed to answer: "given everything I know today, how urgent is this?"

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent cluster-variants <CVE-ID>` — CVE variant clustering

manus-agent cluster-variants CVE-2021-44228
manus-agent cluster-variants CVE-2021-44228 --output json | jq .clusters

Groups CVEs related to the input across three cluster dimensions: same component/vendor, same CWE weakness class, and same researcher/disclosure domain. Useful for finding the full attack surface when one CVE is confirmed exploited.

Flag	Default	Description
`--output {text,json}`	`text`	Output format

`manus-agent changelog` — Manage project changelog

manus-agent changelog                       # show full CHANGELOG.md
manus-agent changelog --version 0.1.0      # show section for a specific version
manus-agent changelog --generate           # preview next release notes from commits
manus-agent changelog --generate --output json

Parses Conventional Commits (feat:, fix:, docs:, etc.) and groups them into Added, Fixed, and Changed sections.

Flag	Default	Description
`--version X.Y.Z`	—	Filter output to a specific release section
`--generate`	off	Preview next release notes from commits since last tag
`--output {text,json}`	`text`	Output format

Configuration

Create ~/.manus-agent/config.toml (or run manus-agent init):

[llm]
provider = "bedrock"          # "openai" | "anthropic" | "bedrock" | "ollama"
model = "us.anthropic.claude-sonnet-4-20250514-v1:0"

[sandbox]
enabled = true
docker_image = "python:3.12-slim"

[tools]
enabled = ["file_ops", "code_execute", "web_search"]

[agent]
# "none" | "sliding_window" | "agentic" (model-managed, recommended for long tasks)
context_manager = "agentic"

See config/config.example.toml for all options.

Provider examples

AWS Bedrock:

[llm]
provider = "bedrock"
model = "us.anthropic.claude-sonnet-4-20250514-v1:0"
# Uses ~/.aws/credentials or IAM role automatically

OpenAI:

[llm]
provider = "openai"
model = "gpt-4o"
api_key = "***"    # or OPENAI_API_KEY env var

Anthropic:

[llm]
provider = "anthropic"
model = "claude-3-5-sonnet-20241022"
api_key = "***"    # or ANTHROPIC_API_KEY env var

Ollama (local):

[llm]
provider = "ollama"
model = "llama3.2"
base_url = "http://localhost:11434"

Python API

# General-purpose agent
from manus_agent import ManusAgent
agent = ManusAgent()
result = agent("Write a Python script that fetches weather data and saves it to CSV")

# Browser automation
from manus_agent.agents import BrowserUseAgent
agent = BrowserUseAgent()
result = agent("Go to GitHub and find the top 5 trending Python repositories today")

# Data analysis
from manus_agent.agents import DataAnalysisAgent
agent = DataAnalysisAgent()
result = agent("Load sales.csv, compute monthly revenue, and plot a bar chart")

# Multi-agent orchestration
from manus_agent.multi_agents import WorkflowAgent
workflow = WorkflowAgent()
result = workflow.handle_request("""
    1. Search the web for recent AI research papers
    2. Analyse the trends and create visualisations
    3. Generate a comprehensive report
""")

# Vulnerability intelligence
from manus_agent.agents import VulnerabilityIntelligenceAgent
from manus_agent.config import Config
agent = VulnerabilityIntelligenceAgent(config=Config.from_file())
result = agent.handle_request("Analyse CVE-2025-6554")

Security & Vulnerability Intelligence

Pipeline

The manus-agent analyze command runs an 8-step pipeline:

NVD + GHSA — official CVE metadata, CVSS, CWE
CISA KEV — known-exploited-vulnerabilities catalogue; VulnCheck KEV when VULNCHECK_API_KEY is set (100+ intel sources, ransomware flag)
AlienVault OTX — threat intelligence pulses and IoCs
PoC discovery — Trickest/CVE index, Exploit-DB, PacketStorm, GitHub, VulnCheck KEV (with API key)
URL verification — every candidate URL is fetched and validated
Deep analysis — patch diff summary · exploit complexity score · version range resolution · vendor response status · PoC freshness
CWE correlation — weakness classification and remediation hints
Report generation — structured text, JSON, or Lark document

Quick reference

# Full intelligence report
manus-agent analyze CVE-2024-3094

# How exploitable is it right now?
manus-agent exploit-complexity CVE-2024-3094
manus-agent epss-trend CVE-2024-3094

# What changed in the fix?
manus-agent patch-diff CVE-2024-3094

# Am I affected?
manus-agent version-range CVE-2024-3094
manus-agent sbom-scan bom.json

# Triage two CVEs
manus-agent compare CVE-2024-3094 CVE-2021-44228
manus-agent temporal-priority CVE-2024-3094

# Find related exposure
manus-agent blast-radius CVE-2021-44228
manus-agent cluster-variants CVE-2021-44228
manus-agent silent-patches apache/log4j

VulnCheck enrichment (optional)

Set VULNCHECK_API_KEY to unlock two additional intel sources:

Index	What it adds
`vulncheck-kev`	Exploitation status from 100+ sources (FBI Flash, CERT advisories, …); prints 🚨 ACTIVELY EXPLOITED banner when confirmed; ⚠️ RANSOMWARE ASSOCIATED when applicable
`nist-nvd2`	Enriched CPE matching with additional CVSS metadata and version ranges

Note: These tools are for defensive security purposes only.

Key Features

Agent types

Agent	Class	Best for
General	`ManusAgent`	File ops, code execution, reasoning
Browser (full JS)	`BrowserUseAgent`	JS-heavy sites, form filling, scraping
Browser (lightweight)	`BrowserAgent`	Static pages, simple navigation
Data analysis	`DataAnalysisAgent`	CSV/JSON processing, charts
MCP	`MCPAgent`	Model Context Protocol tool servers
Multi-agent	`WorkflowAgent`	Complex tasks needing multiple specialists
Vulnerability intel	`VulnerabilityIntelligenceAgent`	CVE analysis, threat intelligence
Remediation	`RemediationAgent`	Actionable fix guidance
Variant analysis	`VariantAnalysisAgent`	Finding similar bugs in related codebases
CVE discovery	`VulnerabilityDiscoveryAgent`	High-EPSS CVE triage and tracking

LLM providers

AWS Bedrock — Claude, Titan, …
OpenAI — GPT-4o, GPT-4-turbo, …
Anthropic — Claude 3.5 Sonnet, Opus, …
Ollama — Llama, Mistral, … (local)

Development

git clone https://github.com/manus-use/manus-agent.git
cd manus-agent
pip install -e ".[dev,browser,search,visualization]"

Run tests

pytest tests/ -v
pytest tests/ --cov=manus_agent --cov-report=html

# Via hatch
hatch run test
hatch run test-cov

900+ tests, all HTTP calls mocked.

Lint and format

ruff check src/ tests/
ruff format src/ tests/

# Via hatch
hatch run lint
hatch run format

Project layout

manus-agent/
├── src/manus_agent/
│   ├── agents/          # Agent implementations
│   │   ├── base.py
│   │   ├── manus.py
│   │   ├── browser.py / browser_use_agent.py
│   │   ├── data_analysis.py
│   │   ├── mcp.py
│   │   ├── vi_agent.py          # VulnerabilityIntelligenceAgent
│   │   ├── remediation_agent.py
│   │   ├── variant_agent.py
│   │   └── vulnerability_discovery_agent.py
│   ├── multi_agents/
│   │   └── workflow_agent.py
│   ├── tools/           # Strands tool implementations (one file per tool)
│   ├── cli.py           # manus-agent CLI entry point
│   ├── config.py
│   └── __init__.py
├── tests/               # pytest test suite (900+ tests, all mocked)
├── config/
│   └── config.example.toml
├── examples/
├── CHANGELOG.md
└── pyproject.toml

See CONTRIBUTING.md for contribution guidelines.

Support

🐛 Issue Tracker
💬 Discussions

License

MIT — see LICENSE for details.

Changelog

See CHANGELOG.md.

Acknowledgments

Built with Strands SDK
Browser automation powered by browser-use

Name		Name	Last commit message	Last commit date
Latest commit History 148 Commits
.github/workflows		.github/workflows
config		config
docs		docs
examples		examples
scripts		scripts
skills		skills
src/manus_agent		src/manus_agent
tests		tests
.env.example		.env.example
.gitignore		.gitignore
.pre-commit-config.yaml		.pre-commit-config.yaml
CHANGELOG.md		CHANGELOG.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md
pyproject.toml		pyproject.toml
va_agent.py		va_agent.py
variant_analysis_agent.py		variant_analysis_agent.py

Folders and files

Latest commit

History

Repository files navigation

manus-agent

Table of Contents

Installation

Quick Start

CLI Reference

manus-agent [task] — Run a task

manus-agent init — Configure credentials

manus-agent doctor — Diagnose your environment

manus-agent history — Browse past runs

manus-agent analyze <CVE-ID> — Vulnerability intelligence

manus-agent remediate <CVE-ID> — Remediation guidance

manus-agent discover — CVE discovery

manus-agent epss-trend <CVE-ID> — EPSS score history

manus-agent patch-diff <CVE-ID> — Patch diff summariser

manus-agent variants <CVE-ID> — Variant analysis

manus-agent compare <CVE-A> <CVE-B> — Side-by-side comparison

manus-agent exploit-complexity <CVE-ID> — Exploit complexity scorer

manus-agent poc-search <CVE-ID> — Multi-source PoC aggregator

manus-agent blast-radius <SPEC> — Dependency blast radius

manus-agent silent-patches <owner/repo> — Silent patch detector

manus-agent cve-timeline <CVE-ID> — CVE timeline

manus-agent version-range <CVE-ID> — Affected version ranges

manus-agent vendor-response <CVE-ID> — Vendor response tracker

manus-agent poc-freshness <CVE-ID> — PoC freshness checker

manus-agent sbom-scan <bom-file> — SBOM scanner

manus-agent temporal-priority <CVE-ID> — Temporal priority scorer

manus-agent cluster-variants <CVE-ID> — CVE variant clustering

manus-agent changelog — Manage project changelog

Configuration

Provider examples

Python API

Security & Vulnerability Intelligence

Pipeline

Quick reference

VulnCheck enrichment (optional)

Key Features

Agent types

LLM providers

Development

Run tests

Lint and format

Project layout

Support

License

Changelog

Acknowledgments

About

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Uh oh!

Contributors

Uh oh!

Languages

`manus-agent [task]` — Run a task

`manus-agent init` — Configure credentials

`manus-agent doctor` — Diagnose your environment

`manus-agent history` — Browse past runs

`manus-agent analyze <CVE-ID>` — Vulnerability intelligence

`manus-agent remediate <CVE-ID>` — Remediation guidance

`manus-agent discover` — CVE discovery

`manus-agent epss-trend <CVE-ID>` — EPSS score history

`manus-agent patch-diff <CVE-ID>` — Patch diff summariser

`manus-agent variants <CVE-ID>` — Variant analysis

`manus-agent compare <CVE-A> <CVE-B>` — Side-by-side comparison

`manus-agent exploit-complexity <CVE-ID>` — Exploit complexity scorer

`manus-agent poc-search <CVE-ID>` — Multi-source PoC aggregator

`manus-agent blast-radius <SPEC>` — Dependency blast radius

`manus-agent silent-patches <owner/repo>` — Silent patch detector

`manus-agent cve-timeline <CVE-ID>` — CVE timeline

`manus-agent version-range <CVE-ID>` — Affected version ranges

`manus-agent vendor-response <CVE-ID>` — Vendor response tracker

`manus-agent poc-freshness <CVE-ID>` — PoC freshness checker

`manus-agent sbom-scan <bom-file>` — SBOM scanner

`manus-agent temporal-priority <CVE-ID>` — Temporal priority scorer

`manus-agent cluster-variants <CVE-ID>` — CVE variant clustering

`manus-agent changelog` — Manage project changelog

Packages