Test suite
277/277passing in ~22s100%02 Quality system
Trust is a
feature.
RepoBrain tests isolated capabilities, adversarial repository changes, and complete agent journeys—then turns the lens on its own codebase.
10/10milestones
evaluated
evaluated
Last verified July 22, 2026 · macOS · Python 3.11
Self-index
113real project filesliveKnowledge graph
5,041nodes + edgeslocalIndex warnings
0across self-indexcleanNo-change writes
0incremental convergencestableCapability scorecard
Measured independently. Proven together.
File scanning
Ignore rules, binary detection, size limits, incremental hashing
Code intelligence
Symbols, imports, calls, tests, and environment reads across 8 languages
Markdown mapping
Paths, unique symbols, provenance, ambiguity rejection
Config tracing
Environment definitions, Compose, Actions, and Kubernetes
Data flow
Endpoint to handler, service, repository, and storage
Impact analysis
Ranked callers, tests, docs, configuration, and blast radius
Layered memory
Agent decisions, assumptions, and open questions — anchored to source and re-verified as it changes
Runtime reconciliation
Flask, Express, and SQLAlchemy convention-based route and table edges
Agent integration
Local MCP server, install-agent, and session-scoped tool discovery
Scale & distribution
1,000+ file performance benchmarks, packaged wheel, self-hosted on its own source
i Progress bars show implementation coverage—not quality scores. Verified capabilities are backed by executable tests.
Adversarial lifecycle
One fact, under pressure.
1
Unique symbolGrounded mention created
2
Collision addedAmbiguous edge removed
3
Collision deletedRelationship restored
4
Code renamedStale fact disappears
5
Docs repairedNew evidence connected
6
No-op indexZero graph writes
Release dimensions
No hidden averages.
CorrectnessPassing
False-positive resistancePassing
Source provenancePassing
Confidence calibrationPassing
Incremental convergencePassing
Malformed-input resiliencePassing
1,000-file performancePassing
Agent context qualityPlanned
Whole-system journeys
The final exam is useful work.
Each milestone contributes focused tests and adversarial mutations. These journeys become the release gate when the systems meet.
Configuration safety
Change a database setting without missing an override.
Capabilities ready · journey test pendingRuntime understanding
Modify an endpoint across its complete data flow.
Capabilities ready · journey test pendingDurable context
Resume a refactor without repeating failed work.
Capabilities ready · journey test pendingProject explanation