Scope the change
Compare committed changes from the merge base to your head commit. Withhold sensitive files and redact recognized credential patterns.
BOUNDED DIFFTHE CONTEXT YOUR PIPELINE IS MISSING
Put every
change in context.
Connect code changes to the requirements behind them. Bring bounded AI review, reference validation, and optional memory into the CI pipeline you already use.
.NET 10 / OpenAI + Claude / Your CI
businessharness / pipelineSAMPLE
$ businessharness review \
--base origin/main --head HEAD
✓ Compare committed changes
✓ Bound and redact diff context
✓ Resolve configured references
REFERENCE / EXAMPLE-42
“Reject expired access tokens.”
● METEvidence in added lines
✓ Gate passedreview.json · review.md · review.sarif
$ businessharness scan \
--base origin/main --head HEAD
✓ Compare committed changes
✓ Scan added lines for credential patterns
✓ Write structured local reports
OFFLINE BY DESIGN
No credentials. No API calls.
● LOCALNot an AI review
✓ Scan completeCredential-pattern checks on this diff
$ businessharness memory-search \
--config C:/trusted/harness.policy.json \
--session pr:123 --query "token expiry"
✓ Select explicitly configured scope
✓ Recall bounded historical context
OPT-IN / POWERED BY MAGICORE
Carry the discussion forward.
● SCOPEDCurrent evidence still required
✓ Context recalledMemory never authorizes approval
Code + requirements + context. One review flow.
YOUR PIPELINE.
NO PLATFORM SWITCH.
GitHub ActionsGitLab CIAzure DevOpsBitbucket Pipelines
01 / A MORE COMPLETE REVIEW
The diff is only
part of the story.
A change can look right and still miss the requirement. BusinessHarness connects the two, with explicit gates between model output and pipeline decisions.
Bring the requirements
Discover references in PR or MR metadata. Resolve Jira tickets and configured JSON sources through trusted routing policy.
REFERENCE VALIDATIONReview with evidence
Use OpenAI or Claude for structured review. Validate findings against added lines and require complete reference assessments.
STRUCTURED OUTPUTMake the gate explicit
Write JSON, Markdown, and SARIF reports. Enable host feedback when you choose. Keep review history with opt-in memory.
YOUR POLICY02 / DELIBERATE BOUNDARIES
Useful intelligence.
Explicit control.
AI review belongs inside your engineering process. BusinessHarness keeps its inputs, permissions, and decisions bounded.
Read the security modelThe model reviews. It doesn’t execute.
No shell, filesystem access, or tools for the review provider. Reviewed repository scripts are never run.
Uncertainty stays visible.
Missing, partial, or unverifiable reference assessments fail the gate. Local findings survive model output.
Feedback is your choice.
Local reports are the default. Optional GitHub and GitLab feedback checks fresh metadata before publishing.
Memory starts switched off.
Enable scoped MAGICore memory explicitly. Historical context cannot establish current correctness or approve a change.
03 / START SMALL
Your first scan.
No API key needed.
Build from source, then scan a local repository with two existing commits. Add an AI provider and reference sources when you’re ready.
Read the setup guideInitial implementation
Build from source today. Packages are not yet published to NuGet, a container registry, or an action marketplace. Repository access may be required.
QUICKSTART / POWERSHELL
git clone --recurse-submodules https://github.com/jihadkhawaja/BusinessHarness.git
cd BusinessHarness
dotnet restore BusinessHarness.slnx --locked-mode
dotnet build BusinessHarness.slnx -c Release --no-restore
dotnet run --project src/BusinessHarness.Cli -- scan `
--repo C:/code/my-project `
--base origin/main --head HEAD `
--output artifacts/scanRequires Git and .NET SDK 10.0.403 or a newer patch in that feature band. Replace the repository path; fetch both commits and their common history.
A FEW PRACTICAL DETAILS
Before you plug in.
Which AI providers can I use?
OpenAI’s Responses API and Claude’s Messages API, using compatible structured-output models you configure. You supply your own API credentials. A ChatGPT or Claude app subscription is not used for API access.
Does my source code leave my environment?
The offline scan makes no provider, host, or reference HTTP calls. An AI review sends bounded, redacted diff context, metadata, and resolved references to your selected provider. Redaction is heuristic; review your organization’s data and retention requirements before enabling it.
Can this replace human review or tests?
No. BusinessHarness supplements code review, tests, SAST, and your existing policy controls. Reference assessments are model judgments about supplied evidence, and some runtime requirements cannot be verified from a diff.
How is BusinessHarness licensed?
Free for individuals using it personally for non-commercial purposes. Commercial use and use by or on behalf of an enterprise or other organization require a separate paid license. Contact Jihad Khawaja for licensing, and see the repository’s full terms.
MAKE CONTEXT PART OF THE PIPELINE.