AI FAQs - Governance, Detection & Trust

What are AI Rule Files?

AI Rule Files are configuration files that provide instructions to AI agents or tools for code generation and modification. They define project-specific structures, coding standards, best practices, and constraints—helping ensure AI output aligns with your organization’s development guidelines.

How does LinearB identify AI-generated code?

LinearB detects AI involvement by analyzing metadata signals in your connected Git repositories and supported AI-tool integrations:

1. Commit co-authors — Identifies AI agents listed in commit messages.
2. AI comments — Detects comments generated by AI agents in pull requests.
3. AI-authored PRs — Finds pull requests opened by known AI agents.
4. AI-tool integrations — Uses supported provider APIs to collect adoption and usage data.

These signals help LinearB identify AI-assisted activity across commits and pull requests without indexing or storing your source code.

How can I attribute Codex-assisted commits?

Codex can add a standard Git co-author trailer to commits it materially contributes to. LinearB can use this commit metadata as a signal when identifying Codex-assisted activity.

Add the following rule to your global ~/.codex/AGENTS.md file:

# Git commits
          
          - Whenever you create a Git commit for changes you materially contributed to, keep my configured Git identity as the primary author.
          - Append exactly one Git trailer after a blank line: `Co-authored-by: Codex <codex@openai.com>`
          - Do not duplicate the trailer. Verify the complete message before finishing.

Start a new Codex Desktop task or CLI session after saving the file so Codex loads the updated global instructions.

Your configured Git identity remains the primary author. The trailer records Codex as a co-author only when Codex materially contributed to the commit.

For more information about global Codex instructions, see Custom instructions with AGENTS.md.

Do you show rejection reasons for AI-suggested code?

No. The AI Insights dashboard displays Acceptance Rates, representing how often AI suggestions are approved. The inverse of this rate represents rejected suggestions.

How does LinearB distinguish between human and AI-written code?

For general detection, LinearB identifies AI involvement through commit messages, co-author tags, AI-authored pull requests, and supported AI-tool integrations.

For LinearB AI Review, the system also tracks suggested fixes in the relevant code areas. When a fix is applied, it contributes to the Acceptance Rate metric.

How is AI usage tied to engineering impact?

LinearB labels PRs co-authored by AI to measure their effect on quality and throughput. These insights can be compared with engineering metrics such as Cycle Time, Deployment Frequency, Change Failure Rate, and Mean Time to Recovery to show how AI adoption influences delivery performance.

How can we measure AI trust and adoption among developers?

Start with a gradual rollout, such as enabling AI PR Descriptions, and collect feedback from developers.

Monitor AI adoption, usage, and Acceptance Rates over time in LinearB. Increasing acceptance rates can indicate growing trust and comfort with AI-assisted development.

Can LinearB backfill AI Insights data?

Backfill availability depends on the connected provider and integration method. Refer to the setup documentation for the relevant integration to confirm its supported historical-data window.

Can we filter LinearB AI Review metrics by team or project?

Currently, you can filter by team, service, repository, and PR label. Filtering by project is on the roadmap.

What does the baseline in the AI Review Dashboard represent?

The baseline shows non-AI-generated code during the selected reporting period. This allows side-by-side comparisons between AI-assisted and manual work.

How can we analyze AI co-author tags from Git history at scale?

LinearB analyzes supported commit co-author metadata to identify and aggregate AI-assisted activity. You can view this data across your organization or filter it by supported dimensions such as team and repository.

How can I fine-tune AI Reviews?

Go to Settings > AI Tools > AI Review and edit your Guidelines.

You can add rules such as “Ignore formatting changes” or “Focus on security improvements.”

Can AI Reviews follow my organization’s coding standards?

Yes. Customize review behavior and criteria in Settings > AI Tools > LinearB AI & Automations > AI Reviews > Edit.

Self-managed users can refer to the Configuring AI Reviews documentation for setup steps.

Can we compare DORA metrics between AI-assisted and non-AI-assisted work?

Yes. LinearB lets you compare supported engineering metrics across AI-assisted and non-AI-assisted work.

• Essentials plan: Cycle Time and Deployment Frequency
• Enterprise plan: All four DORA metrics, including Change Failure Rate and Mean Time to Recovery

Does AI Insights require gitStream?

No. AI Insights can detect supported AI-generated activity without requiring gitStream.

Where do Acceptance Rate and AI-usage numbers come from?

LinearB combines metadata from connected Git repositories with data from supported AI-tool integrations, including GitHub Copilot, Claude, and Codex. Available metrics depend on the connected provider and integration method.

What data does LinearB collect?

LinearB processes metadata signals such as commit authors, PR comments, rule file names, and data provided by connected AI-tool integrations.

Your source code is never indexed or stored.

How secure is this data?

Git metadata processing occurs through your authorized Git provider connections. LinearB does not transfer or store raw source-code content for AI detection.

What tools are available through the MCP server?

LinearB’s MCP server provides structured access to data domains including:

• Engineering metrics and benchmarks
• Repositories and contributors
• Releases and incidents
• Issues and project management entities
• Services, users, teams, and branches
• Pull requests
• AI development tools and AI Analytics

Available tools may evolve as LinearB adds new data and capabilities.

Does MCP enforce permissions or role-based access?

MCP access depends on the authentication method, the connected LinearB organization, and the access available to the authenticated user or API key.

Which AI tools can connect to LinearB’s MCP server?

LinearB supports MCP-compatible clients including Claude Desktop, Claude Code, Codex, Cursor, VS Code, Cline, Continue.dev, Windsurf, and Zed.

OAuth is recommended. Clients that do not support OAuth can use a LinearB API key instead.

For client-specific instructions, see Connect to LinearB MCP Server.

Does MCP only access Git data?

No. In addition to Git and engineering metrics, the MCP server can access supported project management entities, incidents, releases, services, and AI Analytics data available in your LinearB organization.