Components
Eleven components of an agent harness · 65,239 indexed · Stack
Identity
Who the agent is, and how the outside world reaches it
- Listed
- 499
- Ranked
- 8.4%
- Top Score
- 99.9
Pick karpathy-coding-disciplineMemory
What the agent keeps between runs
- Listed
- 36
- Ranked
- 100.0%
- Top Score
- 99.8
Pick letta-ai-lettaSkills
Procedures loaded on demand, one file each
- Listed
- 1,132
- Ranked
- 2.6%
- Top Score
- 99.9
Pick superpowersTools
Terminal, browser and computer commands an agent runs
- Listed
- 57
- Ranked
- 94.7%
- Top Score
- 99.9
Pick playwright-cliHooks
Deterministic code fired on agent lifecycle events
- Listed
- 134
- Ranked
- 9.0%
- Top Score
- 99.1
Pick plannotatorSub-Agents
Delegated workers, each with its own context window
- Listed
- 724
- Ranked
- 0.3%
- Top Score
- 99.7
Pick wshobson-agentsMCPs
Servers exposing tools and data over one protocol
- Listed
- 61,707
- Ranked
- 46.5%
- Top Score
- 99.9
Pick github-mcpDispatch
Routing between agents, plus scheduled and looping runs
- Listed
- 59
- Ranked
- 32.2%
- Top Score
- 99.9
Pick pocketflowEvals
Graded tasks that prove a change helped
- Listed
- 37
- Ranked
- 86.5%
- Top Score
- 99.5
Pick promptfooObservability
Traces, cost and errors from each run
- Listed
- 34
- Ranked
- 76.5%
- Top Score
- 99.7
Pick langfuseSandbox
Isolated compute an agent executes and deploys in
- Listed
- 26
- Ranked
- 53.8%
- Top Score
- 99.9
Pick e2b-sandboxProperties
Qualities of the whole harness, not components
- Self-Improvement
Improves itself between runs
- Model Routing
Which model runs which step
- Self-Improvement
Stack names one pick per component. Leaderboard ranks the catalog without the grouping.