Agentproof — Testing skill for Claude Code
Local-first Go CLI that turns coding-agent sessions into integrity-checked merge evidence: Git association, test ingestion, blast radius, and deterministic risk findings.
How to install Agentproof
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open ralabarta/agentproof and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Agentproof does
Local-first Go CLI that turns coding-agent sessions into integrity-checked merge evidence: Git association, test ingestion, blast radius, and deterministic risk findings. No account, no telemetry, stdlib only.
Alternatives in Testing
- Pua — Use when the user invokes /pua or asks for PUA mode, try-harder/retry help, change-approach coaching, completi 19.5k ★
- Testing Strategy — This project is small, runs in a terminal, and is mostly deterministic 10.9k ★
- Baro — A CLI that turns a goal into a pull request - and a sandbox for testing concurrent AI coding agents on the Moz 120 ★
README
AgentProof
**Know exactly what your coding agent changed — and collect the evidence needed to decide whether it is safe to merge.**
[](https://github.com/ralabarta/agentproof/actions/workflows/ci.yml) [](LICENSE) [](https://go.dev) [](#contributing) [](#privacy-and-trust-model)
[Quickstart](#quickstart) · [Why](#why-agentproof) · [GitHub Action](#github-action) · [Real report](docs/example-report.md) · [Architecture](docs/architecture.md) · [Threat model](docs/threat-model.md)
AgentProof is a **local-first Go CLI** that associates Codex and Claude Code sessions with Git changes, ingests test-result artifacts, detects deterministic risks, estimates code impact, and emits reproducible **Markdown, HTML, and JSON evidence**.
No account. No service. No telemetry. No network request.
Quickstart
go install github.com/ralabarta/agentproof/cmd/agentproof@latest
agentproof init
agentproof record --objective "Protect refresh tokens from replay" --agent codex -- codex
agentproof verify --test-result test-results.jsonl
AgentProof verification: WARNING
✓ Required evidence complete: 3/3
✓ Canonical manifest integrity passed
✓ 28 test results passed
✓ No secret patterns detected in captured added lines
⚠ Authentication or authorization code modified
Affected components: internal/auth, internal/api
Bundle ID: 7f0c…d91a
[!IMPORTANT] AgentProof does **not** certify code as safe. It separates observed evidence, deterministic derivations, associations, unsupported checks, and unknowns so reviewers can make a better merge decision.
Related Skills
Ix Impact
Change risk analysis — blast radius, affected systems, and what to test. Depth scales with risk level; low-ris
Decision Memo
Produce a senior-leadership decision memo before proposing a non-trivial change. Seven sections — observation,
Review Impact
Code review with blast-radius analysis. Runs architect (graph trace) + qa-engineer (quality) IN PARALLEL.
Test Resume
A/B verification harness for the resume subagent. Scoring is deterministic via mcpjarvispick_resume_sessions,
Ke Test
Select and run risk-based verification checks, then record exact evidence and remaining uncertainty.
Reviewer Coverage
You are the test-coverage specialist of a pull-request review pipeline. You see one deterministic group of cha
Related Agents
AIOps Manager
Owns AIOps — anomaly detection and correlation over observability telemetry, alert-noise reduction (grouping,
Code Reviewer Bug Hunter
Bug Hunter persona for code review. Focuses on state transitions (cross-module), exception path integrity, dep
Tp Review Impact
PROACTIVELY use this when the user asks "will this PR break production", "what's the blast radius of these cha