rajkumar-pendyala

Sandbox Mini Harness

Development community

SOURCE https://github.com/rajkumar-pendyala/sandbox-mini-harness

Description

A small agent on the Claude Agent SDK that pulls a free public source through an MCP connector you write, scores or analyzes it, drafts an output, routes it through a human-approval step, and logs the run

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

Sandbox mini-harness

Five things, now with step 2 split into a 2-call chain:

Pulls a free public source via an MCP connector we wrote (mcp_server.py, wraps Grants.gov's no-auth search2)
Scores/analyzes it — chain step 1, run_scoring_step()
Drafts an output — chain step 2, run_drafting_step(), fed only step 1's result, no connector needed
Routes it through a human-approval step — human_approval_step(), now also captures a named approver
Logs the run (run_log.jsonl) — one entry per chain step plus the approval decision

No debra data or assets anywhere in this repo.

Setup

pip install claude-agent-sdk mcp httpx
export ANTHROPIC_API_KEY="your-key"   # set in your shell, never in this repo

Run

python3 sandbox_agent.py

You'll see the agent search Grants.gov, then a draft printed to the terminal, then a `y/n` prompt. Nothing sends anywhere either way — it just logs your decision.

Check the log

cat run_log.jsonl