Pipeline Minder

The Pipeline Minder is Maudel’s real-time quality supervision system. It monitors every pipeline execution with 12 pluggable evaluators, automatically detecting and remediating issues before they compound.

12 Pluggable Evaluators

The Pipeline Minder covers the full spectrum of execution quality:

  • Worktree isolation — ensures all work happens in sandboxed git worktrees
  • Git state — monitors branch state, uncommitted changes, and merge readiness
  • Build awareness — tracks build status and catches compilation errors early
  • Input quality — validates that stage inputs meet expected schemas and completeness
  • Output quality — evaluates generated artifacts against rubric criteria
  • Playwright enforcement — ensures browser-based tests execute correctly
  • Config safety — prevents accidental modification of protected configuration
  • Port safety — protects reserved system ports from being claimed by agents
  • Scope compliance — detects when agent work drifts beyond the assigned task scope
  • Merge verification — validates that worktree changes merge cleanly
  • Dependency checking — ensures all task dependencies are satisfied before advancing
  • Pattern detection — identifies recurring issues across executions

Severity-Gated Auto-Remediation

Not all issues require human intervention. The Pipeline Minder uses severity levels to determine the appropriate response:

  • Critical — auto-remediate immediately (e.g., git stash, commit recovery, worktree reset)
  • High — flag for review with recommended fix
  • Medium — add to findings queue for end-of-run report
  • Low — log for pattern analysis

Cross-Execution Pattern Detection

The Pipeline Minder tracks issues across multiple executions, identifying recurring patterns that indicate systemic problems rather than one-off failures. This feeds into the Learning System for long-term improvement.

Live Dashboard

Real-time visibility into pipeline health:

  • Findings feed — live stream of evaluator results
  • Review queue — issues requiring human attention
  • End-of-run report — comprehensive summary of all findings, remediations, and recommendations
  • Historical trends — track quality metrics across executions

Ready to transform your AI engineering?

Governed AI engineering, from idea to deployment.