What's New in OpenClaw 3.24: Skills, Teams, and Sub-Agents

OpenClaw 3.24 shipped this week with four headline features: native teams support, first-party Slack integration, a rebuilt sub-agent system, and new APIs for skill developers. This is the most significant release since 3.20 introduced the plugin framework, and it changes how teams collaborate with AI agents and how skill authors build for the platform.

Here is what is new, what changed, and what it means for your workflows.

Native Teams Support

OpenClaw has supported individual developer workflows since launch. Version 3.24 adds first-class support for teams — shared configurations, centralized skill management, and collaborative agent sessions.

Shared Skill Configurations

Teams can now define a shared skill set at the organization level. Instead of every developer installing and configuring skills individually, a team admin creates a skill profile that applies to all members:

# .openclaw/team-skills.toml
[team]
name = "frontend-squad"

[[skills]]
name = "nextjs-app-router-expert"
version = "2.1.0"
required = true

[[skills]]
name = "typescript-strict-patterns"
version = "1.8.3"
required = true

[[skills]]
name = "tailwind-v4-utilities"
version = "1.2.0"
required = false

When a team member runs openclaw sync, they get the full skill set automatically. Required skills are always loaded; optional skills can be toggled per developer. This eliminates the "works on my machine" problem for agent-assisted development — every team member's agent has the same base knowledge.

Centralized Skill Pinning

Team admins can pin specific skill versions to prevent uncoordinated upgrades. This is critical for large teams where a skill update could change the agent's behavior mid-sprint. Pin the version in the team config, test the upgrade in a branch, and roll it out when the team is ready.

Usage Analytics

The teams dashboard provides aggregate analytics: which skills are used most, which generate the most tool calls, and where agents spend the most tokens. This data helps teams optimize their skill stack and identify opportunities for custom skills that address repeated patterns.

Slack Integration

OpenClaw 3.24 includes a first-party Slack integration that connects your agent to team communication channels. This is not a simple notification bot — it is a full bidirectional bridge that lets your agent participate in Slack conversations.

How It Works

Install the Slack integration skill from the Bazaar:

openclaw skill install slack-integration

Configure it with your Slack workspace credentials:

openclaw config set slack.workspace_id "T0123ABCD"
openclaw config set slack.bot_token "xoxb-..."

Once connected, your agent can:

• Read channel messages to understand team context and ongoing discussions
• Post updates when tasks complete, builds finish, or reviews are ready
• Respond to mentions — tag the agent in a channel and it responds with full context awareness
• Thread participation — the agent follows conversation threads and provides relevant input

Use Cases for Teams

The Slack integration shines in team workflows:

• Standup automation: The agent reads the standup channel, summarizes what each team member posted, and flags blockers that need attention
• Code review notifications: When the agent finishes reviewing a pull request, it posts a summary in the team's review channel with key findings
• Incident response: During an outage, the agent monitors the incident channel and pulls relevant logs, recent deployments, and similar past incidents into the conversation
• Knowledge sharing: Team members ask questions in a channel and the agent answers using the team's codebase, documentation, and installed skills as context

Privacy and Permissions

The integration respects Slack's permission model. The agent only accesses channels it has been explicitly invited to. Direct messages are never read unless the user initiates a conversation with the agent bot. All data flows through your workspace's existing security policies.

Sub-Agent Improvements

Sub-agents — secondary agents that the primary agent spawns to handle specific tasks — got a major overhaul in 3.24. The previous implementation had limitations around context sharing, error handling, and skill inheritance that made sub-agents unreliable for complex workflows.

What Changed

Context isolation with selective sharing: Sub-agents now run in isolated context windows by default. The parent agent explicitly passes relevant context rather than sharing its entire state. This prevents sub-agents from being overwhelmed by irrelevant context and reduces token consumption.

spawn sub-agent "test-writer" with context:
  - file: src/auth/login.ts
  - file: src/auth/login.test.ts
  - skill: jest-testing-patterns
  - instruction: "Write additional edge case tests for the login function"

Skill inheritance controls: Previously, sub-agents inherited all of the parent's skills, which sometimes caused conflicts. Now you can specify which skills a sub-agent inherits:

[sub-agent.test-writer]
inherit_skills = ["jest-testing-patterns", "typescript-strict-patterns"]
exclude_skills = ["code-review-checklist"]

Improved error recovery: When a sub-agent encounters an error, it now reports back to the parent agent with structured error information instead of failing silently. The parent can retry, adjust parameters, or fall back to handling the task itself.

Parallel execution: Multiple sub-agents can now run in parallel with proper resource management. The runtime allocates token budgets per sub-agent and prevents any single sub-agent from consuming the entire allocation.

What This Means for Skill Developers

If you build skills that are intended for use with sub-agents, version 3.24 introduces new metadata fields:

[skill.metadata]
sub_agent_compatible = true
recommended_context_size = "small"
parallel_safe = true

These fields help the runtime make better decisions about how to allocate resources and which skills to load when spawning sub-agents. Update your skills in the Bazaar to include this metadata — it will improve the experience for users on 3.24 and later.

New Skill APIs

Version 3.24 introduces several new APIs that skill developers have been requesting:

Conditional Skill Loading

Skills can now define conditions under which they should be loaded. Instead of loading every installed skill at session start, the runtime evaluates conditions and only loads relevant skills:

[skill.conditions]
file_patterns = ["*.test.ts", "*.spec.ts"]
directories = ["__tests__", "tests"]

This skill only loads when the agent is working in or near test files. Conditional loading reduces context window usage and improves response quality by ensuring the agent is not distracted by irrelevant instructions.

Skill Composition

Skills can now declare dependencies on other skills:

[skill.dependencies]
requires = ["typescript-strict-patterns >= 1.5.0"]
conflicts = ["typescript-loose-mode"]
enhances = ["nextjs-app-router-expert"]

The runtime resolves these dependencies automatically. If you install a skill that requires another skill, the missing dependency is installed for you. Conflicting skills produce a warning at install time rather than causing unpredictable behavior at runtime.

Structured Output Templates

Skills can now define output templates that structure the agent's responses for specific tasks:

[skill.templates.code_review]
format = """
## Summary
{{summary}}

## Issues Found
{{#each issues}}
- **{{severity}}**: {{description}} ({{file}}:{{line}})
{{/each}}

## Recommendations
{{recommendations}}
"""

When the agent uses this skill for a code review, the output follows the template structure. This is particularly useful for teams that want consistent, predictable formats for reviews, documentation, and reports.

Migration Guide

Upgrading to 3.24 is straightforward:

openclaw update

Existing skills and configurations continue to work without changes. The new features are opt-in:

• Teams: Create a .openclaw/team-skills.toml file to start using shared configurations
• Slack: Install the slack-integration skill and configure your workspace credentials
• Sub-agents: Existing sub-agent workflows work as before; new isolation and inheritance controls are available when you update your configurations
• Skill APIs: Existing skills do not need to add the new metadata fields, but doing so improves the experience on 3.24

What Is Next

The 3.25 release, expected in late April, will focus on the MCP server ecosystem. The team is working on a standardized MCP server packaging format that makes it easier to publish and install MCP servers through the Bazaar, along with a local MCP server runtime that eliminates the need for Docker or external hosting for development use.

Follow the OpenClaw Bazaar blog for release updates, and browse the skills directory to find skills that take advantage of the new 3.24 features.

Browse the Skills Directory

Find the right skill for your workflow. The OpenClaw Bazaar skills directory has over 2,300 community-rated skills — searchable, sortable, and free to install.

Browse Skills →

Try a Pre-Built Persona

Don't want to configure everything from scratch? OpenClaw personas come pre-loaded with skills, memory templates, and workflows designed for specific roles. Compare personas →