bharat3645/The-Ideal-Harness
🏪Marketplacebharat3645/The-Ideal-Harness
|A Claude Code-native agent harness with an MCP-portable enforcement and tooling core. Clean-room synthesis of best-of-breed harness ideas — zero overlap, safety enforced below the LLM.
Add this Marketplace
/plugin marketplace add bharat3645/The-Ideal-HarnessPlugins in this Marketplace
ideal-harness-compress
Deterministic, cache-safe tool_result compression (JSON sampling, log RLE, stack-trace collapse) with a Compress-Cache-Retrieve store, plus the caveman output-side terse mode.
ideal-harness-core
Substrate: plugin loader, skill templating + multi-host generation, validation, bootstrap skill. Required by all other modules.
ideal-harness-guard
Enforcement floor below the LLM: deny-wins policy, permission defaults, injection defense, secret redaction + broker, skill vetting, drift-guard, sandbox.
ideal-harness-memory
Structural code-graph with token-budgeted subgraph retrieval + episodic memory with BM25 relevance ranking and a curator that reconciles claims against tool-call evidence.
ideal-harness-orchestrate
Orchestration & control flow: durable task ledger, tool registry, loop/no-progress guard, spend governor, API retry/backoff, session resume/checkpoint, plus subagent-driven-development and brainstorming skills.