Clean Slate - Setup Guide¶
Source: sidhartha1s/clean-slate Category: Agent Infrastructure / Quality Assurance License: MIT · Published: June 28, 2026
Every coding agent eventually says "this should work" when it doesn't. Clean Slate is a session-end verification skill that audits every claim - merged PRs, passing tests, fixed bugs - against the real artifact (git history, CI status, test output). It reconciles unmerged work, flags half-finished changes, and produces a verifiable status report.
The Problem It Solves¶
| Agent Claims | Clean Slate Verifies |
|---|---|
| "PR merged" | Checks git log and remote for the actual merge commit |
| "Tests pass" | Runs the test suite and captures exit codes |
| "Bug fixed" | Searches for the fix commit, verifies it addresses the report |
| "Branch cleaned up" | Lists stale branches, flags unpushed work |
| "Documentation updated" | Diffs docs/ against last tag |
Installation¶
npx skills add sidhartha1s/clean-slate
Direct clone:
git clone https://github.com/sidhartha1s/clean-slate.git
cp clean-slate/SKILL.md ~/.hermes/profiles/<your-profile>/skills/clean-slate.md
Usage with Hermes Agent¶
Invoke at the end of any coding session:
"Run clean-slate verification on this session"
Or trigger automatically by adding to your session-end workflow:
# ~/.hermes/config.yaml
hooks:
session_end:
- skill: clean-slate
args: --strict
Output Example¶
Clean Slate Report - Session #8472
===================================
✅ PR #342 merged - commit a3f2b1c on main
✅ Tests pass - 142/142, 0 failures
⚠️ Branch fix/login-timeout not deleted
❌ Bug #591 claim "fixed" - no related commit found
⚠️ 2 files modified but not committed:
- src/auth/session.ts
- docs/CHANGELOG.md
Verdict: SESSION NOT CLEAN - 1 unverified fix, 2 uncommitted files
CorpusIQ Use Cases¶
| Use Case | How |
|---|---|
| Agent session hygiene | Run at end of every coding session to catch loose ends |
| CI/CD gate | Add as pre-merge check - block PRs from unclean sessions |
| Multi-agent coordination | Verify one agent's claims before another agent builds on them |
| Onboarding QA | New contributors run clean-slate to build verification habits |
This guide is part of the Hermes Skills Catalog. Discovered June 29, 2026. Powered by CorpusIQ.
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.