The complete AI pipeline
Six stages. Full transparency. From ticket to merged PR with safety at every step.
Ticket Ingestion
Your ticket arrives from any source
EnsureFix connects to your existing ticket systems via webhooks or polling. When a new ticket is created or tagged, it automatically enters the processing queue with priority scoring.
- Supports Jira, Azure DevOps, GitHub Issues, Bitbucket
- Webhook-based for instant pickup or configurable polling
- Priority queue with aging bonus for older tickets
- Ticket similarity detection groups related issues
- Batch scheduling for efficient multi-ticket processing
Intelligent Planning
AI analyzes your codebase and plans the fix
The PlannerAgent reads the ticket description, scans your repository tree, and produces a detailed implementation plan listing exact files to modify with per-file intent descriptions.
- Hybrid file ranking: dependency graph (40%) + semantic search (40%) + code similarity (20%)
- Root cause analysis identifies the actual problem, not just symptoms
- Impact simulation models behavioral changes before code is written
- Repo Intelligence Layer validates plan against custom rules
- Plan quality guard catches architectural issues before code gen
- Optional human approval gate before proceeding
Code Generation
Production-ready code, written and reviewed by AI
The CoderAgent generates code in intelligent batches, with each batch reviewed by the ReviewerAgent and SecurityAgent. Self-healing loops automatically fix test failures.
- Intelligent batching: max 5 files per batch, 12 per run
- ReviewerAgent checks: logic errors, security, breaking changes, N+1 queries
- SecurityAgent scans for OWASP vulnerabilities
- Self-healing: detects test failures and auto-fixes without human intervention
- 16-point post-generation validation (behavior mismatch, regression risk, etc.)
- Decision engine: auto_apply / needs_review / block based on confidence
Review & Approval
Human-in-the-loop when it matters
High-confidence fixes can auto-apply. Complex changes surface for human review with full reasoning traces, diff views, and confidence breakdowns.
- Trust panel with confidence ring, pattern match badge, risk breakdown
- Expandable reasoning timeline: 7 layers of AI decision-making
- Inline syntax-highlighted diff viewer
- Safety gate requires acknowledgment for blocked fixes
- Structured refinement: rejected fixes get targeted improvement prompts
- Fix feedback loop trains the learning engine
Commit & Deploy
Branch pushed, PR opened, CI monitored
EnsureFix pushes a branch to your repository, creates a pull request with full context, and monitors CI. If builds fail, the AI automatically diagnoses and fixes.
- Branch creation with configurable naming conventions
- PR description includes ticket context, reasoning, and risk assessment
- CI/CD failure auto-diagnosis via CIFeedbackAgent
- Automatic re-push after CI fixes
- Commit policy enforcement (max files, risk level, blocked paths)
- Auto-merge option for approved, low-risk changes
Learn & Improve
Every outcome makes the system smarter
When you accept or reject a fix, EnsureFix learns. The self-improving engine calibrates weights, extracts patterns, and blocks approaches that repeatedly fail.
- Pattern learning: identifies successful code patterns from accepted fixes
- Weight calibration: per-signal rejection rates inform future scoring
- Failure memory: blocks patterns with 70%+ rejection rate
- 3-tier contextual weights: repo-specific → problem-type → global
- Strategy boosting: successful strategies get confidence bonus
- Reasoning pattern store with Jaccard similarity matching
Ready to automate your pipeline?
See EnsureFix process a real ticket from your backlog in a live demo.