Feature Parity: Bridge Gap with Claude Code Capabilities

[simonrosenberg]

Feature Parity: Claude Code Capabilities Roadmap

This is the master tracking issue for achieving feature parity with Claude Code capabilities in the OpenHands software-agent-sdk. Gaps were identified across 8 areas and organized into epics and tracking issues.

Epics (with sub-issues)

• [Epic] Skills: Execution isolation, model routing & enhanced capabilities #2053 — Skills: Subagent execution, model routing & enhanced capabilities (5 sub-issues)
• [Epic] Delegation: Rewrite APIs, add file-based sub-agents and implement Explore & Bash Agents #2054 — Delegation: Background execution, model routing & file-based agents (7 sub-issues)

Tracking Issues (gaps listed inline)

• [Tracking] MCP: Advanced integration gaps [WORK IN PROGRESS] #2036 — MCP: Advanced integration gaps (7 items)
• [Tracking] Memory: Persistent auto-learning across sessions #2037 — Memory: Persistent auto-learning across sessions
• [Tracking] Session Management & Checkpointing gaps #2038 — Session: Checkpointing, selective undo & CLI management (3 items)
• [Tracking] Hooks: Advanced event system gaps #2039 — Hooks: Advanced event system gaps (6 items)
• [Tracking] Task Tracker: Dependencies, IDs & operations gaps #2040 — Task Tracker: Dependencies, IDs & operations gaps (4 items)
• [Tracking] Tools: Missing tool types & dynamic discovery gaps #2041 — Tools: Dynamic discovery & structured user clarification gaps (2 items)
• [Tracking] Self-Knowledge: Built-in agent guide & skill authoring assistance #2065 — Self-Knowledge: Built-in agent guide & skill authoring assistance

Prioritization

High-impact areas to consider first:

1. Delegation ([Epic] Delegation: Rewrite APIs, add file-based sub-agents and implement Explore & Bash Agents #2054) — Enables parallel agent workflows, critical for complex tasks. Skills isolation (feat(skills): Run skills via subagent delegation for isolated execution #2042) also depends on the delegation system.
2. Skills ([Epic] Skills: Execution isolation, model routing & enhanced capabilities #2053) — Subagent execution is foundational for reliable skill execution
3. Self-Knowledge ([Tracking] Self-Knowledge: Built-in agent guide & skill authoring assistance #2065) — Agent should be self-documenting, especially for skill authoring
4. MCP ([Tracking] MCP: Advanced integration gaps [WORK IN PROGRESS] #2036) — Broad ecosystem integration
5. Memory ([Tracking] Memory: Persistent auto-learning across sessions #2037) — Self-improving agent across sessions
6. Session ([Tracking] Session Management & Checkpointing gaps #2038) — Checkpointing and CLI management
7. Hooks ([Tracking] Hooks: Advanced event system gaps #2039) — Advanced event system

[enyst]

Status update on plugin/AgentSkills parity:

• PR feat(plugin): Add installed plugins management utilities #2031 adds installed plugin management utilities (install/uninstall/list/get/load/update) for Claude-style plugin directories under ~/.openhands/plugins/installed/.
• There is still no enable/disable state in the SDK; enablement happens by passing selected plugins into Conversation/AgentContext.
• AgentSkills are still loaded via load_skills_from_dir / load_user_skills / load_public_skills only; no install/uninstall/list/enable/disable API yet. I opened AgentSkills management utilities #2301 to track AgentSkills management utilities.