Work
Case Study
AgentReceipt
Local-first Go CLI for recording AI coding sessions and producing verifier-ready replay evidence.
- Go
- CLI
- Ed25519
- Developer Infrastructure
- Observability
- Replay
Overview
AgentReceipt is a local-first CLI for AI coding sessions. It records developer-agent activity, signs receipts, and produces replay/focus reports that downstream reviewers or coding-agent loops can consume.
Problem
AI-assisted software work is hard to review when git state, filesystem changes, instruction context, provider logs, commands, quality checks, and final patch evidence are scattered across local tools.
My role
I designed and built the CLI, evidence model, local capture workflow, receipt signing, replay/focus contracts, and verifier-facing review outputs.
Technical details
- Captures git snapshots and diffs, filesystem watcher events, instruction-file metadata, and best-effort Codex or Claude provider evidence.
- Exports signed receipts, portable replay bundles, PR-ready review artifacts, and machine-readable replay/focus JSON for verifier workflows.
- Reports quality gates, failed commands, patch summaries, policy checks, privacy metadata, claims, outcomes, and ranked focus tasks for agent-friendly review loops.
Hard parts and tradeoffs
- The CLI needs enough evidence for independent review while staying local-only and avoiding prompt upload by default.
- Replay and focus outputs need stable contracts for automation without turning the tool into an agent scorer, policy engine, or orchestrator.
Current state
The project is public on GitHub with a tagged v0.9.0 release.