AagamS06

Agent Receipts — Development skill for Claude Code

Development community

Turn Claude Code & Codex session logs into reviewable receipts for agent-authored PRs: verification verdict, risky actions, tools, tokens.

How to install Agent Receipts

This entry records only its repository, not the path inside it, so there is no exact command to give. Open AagamS06/agent-receipts and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What Agent Receipts does

Turn Claude Code & Codex session logs into reviewable receipts for agent-authored PRs: verification verdict, risky actions, tools, tokens.

Alternatives in Development

Compile Knowledge — Compile literature, meeting notes, session logs, and code findings into 130 ★
Agenfk Calc Tokens — Calculate and log token usage for the current session 64 ★
Partner Skill — 搭子.skill / Partner：Claude Code 负责规划、润色和审查，Codex 负责实现、监控和验收 62 ★

README

🧾 agent-receipts

**Show your work, agent.** A local-first CLI + GitHub Action that turns Claude Code and Codex CLI session logs into a reviewable *receipt* for every agent-authored pull request.

[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE) ![Python](https://img.shields.io/badge/python-3.11%2B-3776AB) ![Status](https://img.shields.io/badge/status-phase%200%20%E2%80%94%20scaffold-orange) ![Local-first](https://img.shields.io/badge/data-stays%20local-2B8A3E)

The problem

Coding agents now write code faster than humans can review it. When an agent opens a PR, the reviewer sees the final diff — but not **what the agent ran, whether tests ran on the final code, or whether it did something risky along the way**. That evidence already exists in local JSONL session logs; nobody turns it into something a reviewer can read in a minute.

agent-receipts reads those logs and answers four questions per PR:

Was it verified? VERIFIED · STALE (tests ran before the last edit) · FAILED · UNVERIFIED
Did anything risky happen? rm -rf, force-push, .env reads, curl | sh, DB drops, CI edits…
What did it assume? Agent-stated assumptions, pulled out of the transcript.
What did it use and cost? Tools, MCP servers and tokens across Claude Code and Codex.

Who it helps

Solo devs and small teams who let agents open PRs, reviewers who want a consistent evidence bar, and OSS maintainers receiving agent-generated contributions.

Where the idea came from

Built from a gap analysis of YC **Fall 2026** and **Summer 2026** companies. The same pain points showed up again and again in their public descriptions: the verification bottleneck (Workers IO, Simulithic, GitCafe), silent destructive actions (Agent FM, Decawork, machine0, OneCLI), no cross-harness view of tool usage (Codag, Executor), babysitting many agents (Herdr, Hoplite) and debugging AI-written code (HyperProbe). This is an independent student p