Tried sponsors/affaan-m: A Practical Agent Harness for Claude Code Workflows

affaan-m is gaining serious attention today with +523 GitHub stars, and the reason is easy to understand: it treats AI coding agents as systems that need operational discipline, not just better prompts.

The project focuses on an agent harness for Claude Code, Codex, OpenCode, Cursor, and similar tools, combining reusable skills, behavioral instincts, memory, security practices, and research-first development. In practice, that means creating a repeatable workflow around an agent instead of starting every task from an empty chat window.

The most interesting angle for indie hackers is the potential ROI. A good harness can reduce context thrashing, repeated explanations, unsafe commands, and wasted model calls. That matters when running agents inside Docker or CI pipelines where every request affects both build speed and monthly API spend.

My primary gateway test would use claude-fable-5 through an OpenAI-compatible endpoint:

export OPENAI_API_KEY="your-b-lost-key"
export OPENAI_BASE_URL="https://b-lost.com/v1"
export OPENAI_MODEL="claude-fable-5"

curl "$OPENAI_BASE_URL/chat/completions" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-fable-5",
    "messages": [
      {
        "role": "system",
        "content": "Follow the project harness: research first, verify changes, and minimize risky commands."
      },
      {
        "role": "user",
        "content": "Inspect this repository and propose the smallest safe implementation plan."
      }
    ],
    "temperature": 0.2
  }'

The same base URL can be configured in Cursor, Aider, Cline, Roo Code, Windsurf, NextChat, or LibreChat. B-Lost’s Universal Relay Core advertises 20% off list pricing, plus native Anthropic prompt caching with 90% discounts on cache hits. For long-lived agent sessions, caching can improve both latency and cost—especially when the harness keeps a stable system prompt, skills library, and repository instructions.

This is less about a flashy demo and more about building an agent operating system that ships reliably. That is probably why the star count is moving so quickly.