SatyamSingh-Git/ElevenPowers

Your coding agent says it's done. This asks for the receipts.
**The idea** · **The survey** · **What's different** · **In your project** · **Install** · **Status** · **Credit** · **The journey**
**🗺️ Open the live architecture graph** — five views, including a contributor roadmap
---
> **What this is.** An evidence and verification layer for coding agents, with change-impact analysis and project health checks.
> It watches the commands your agent already runs, turns their output into **evidence**, and binds each record to the exact bytes of the files it observed.
> It helps identify what a change may affect, keeps earlier behavior checks visible across tasks, and **computes a verification state** when the agent claims it's done.
> Edit an observed file afterwards and its green tests go **stale** — the way make invalidates an object file. Missing, failing and incomplete evidence stays explicit; choose reporting or completion gating through the project profile.
**What it can do today:**
1. **ImpactGraph:** maps a change's possible blast radius to affected files, dependencies and candidate tests, grounding plans and milestone recheck advice in repository evidence while showing coverage gaps.
2. **Milestone verification:** tracks checks for previously built behavior across tasks, flags failed, stale, missing or incomplete evidence, and can advise which checks to rerun as development continues.
3. **Test strength:** optionally makes small mutations to edited production code in isolation and checks whether passing tests detect them, exposing possible gaps in regression protection.
4. **Project readiness and health:** discovers project verification commands and checks whether the agent integration is capturing edits, command results and completion correctly, surfacing setup and coverage problems.
5. **Resumable verification:** saves check progress and receipts so interrupted work can resume, fresh results can be reused and later edits can invalidate outdated evidence.
6. **Evidence reports:** exports Markdown or JSON showing what was checked, whether source and selected approved dependencies still qualify the results, where evidence is missing and what needs attention next.
7. **PatchProof:** plans and runs declared checks, binds patches and command-scoped approved inputs to qualified evidence, and shares impact, compatibility and verification gaps through local bundles and GitHub summaries.
8. **Five host integrations:** Claude Code, Codex, Gemini CLI, Cursor Agent and GitHub Copilot CLI.
---
*The path is ours to walk.* *Where it ends was never ours to choose.*
**So walk it honestly — and keep the receipts.**
---
The thirty-second version
Ask a coding agent to fix a bug and it will tell you the bug is fixed.
It is telling the truth as it understands it. That is the problem. Published measurements put the gap between *claiming* completion and *achieving* it in the tens of percent — one study found a frontier model that submitted a patch on every single run and resolved 44% of them. Prompting the model to be more careful does not close that gap, because the model was never lying. It simply has no way to know.
So stop asking it.
Every agent already runs commands. Tests, typechecks, builds, the failing reproduction it wrote three minutes ago. All of that output is flowing past unread on its way to the scrollback. ElevenPowers reads it, files it as evidence, and stamps each record with a fingerprint of the files it saw.
Then "done" stops being a sentence the model emits and becomes a state you can compute:
And because each record is bound to file content, evidence rots when the code moves under it:
The closest analogue is not another agent framework. It is make.
| State | Meaning |
|---|---|
| VERIFIED | every obligation met by fresh evidence |
| UNVERIFIED | an obligation has no evidence |
| STALE | evidence exists, but the files it observed have changed |
| CONTRADICTED | the latest evidence for something is failing |
A claim moves between them on its own, as the code moves:
> [!NOTE]
> A failure that was later fixed is *reproduction evidence*, not a contradiction — that's the point of demanding the test failed first. Only the most recent record per identity counts.
---
First, I went and read the competition
Before writing a line of runtime, I cloned fourteen agent systems, pinned each to a commit, and read them from source. Not the READMEs — the source. Where the loop exits. What is actually *enforced in code* versus what is merely *requested in a prompt*. Where each one is genuinely brilliant, and where it quietly gives up.