Files
git.stella-ops.org/docs/features/dropped/playbook-learning.md

38 lines
2.8 KiB
Markdown

# Playbook Learning (Run-to-Patch Pipeline)
## Module
AdvisoryAI
## Status
PARTIALLY_IMPLEMENTED
## Description
Run artifacts and evidence bundles support playbook-related data, but dedicated playbook learning, patch proposal generation, and versioned playbook management are not fully distinct modules yet.
## What's Implemented
- **Run tracking infrastructure**: `RunService` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Runs/RunService.cs`) tracks runs with artifacts and events
- **Run models**: `Run`, `RunArtifact`, `RunEvent` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Runs/Models/`) capture run outcomes
- **Run storage**: `InMemoryRunStore` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Runs/InMemoryRunStore.cs`) persists run data
- **Evidence bundle assembly**: `EvidenceBundleAssembler` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Chat/Assembly/EvidenceBundleAssembler.cs`) assembles evidence packs from data providers
- **Remediation planning**: `AiRemediationPlanner` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Remediation/AiRemediationPlanner.cs`) generates fix plans
- **PR generation**: `GitHubPullRequestGenerator`, `GitLabMergeRequestGenerator`, `AzureDevOpsPullRequestGenerator` create PRs from remediation plans
- **Run API endpoints**: `RunEndpoints` (`src/AdvisoryAi/StellaOps.AdvisoryAI.WebService/Endpoints/RunEndpoints.cs`) exposes run data
- **Advisory output persistence**: `AdvisoryOutputStore` (`src/AdvisoryAi/StellaOps.AdvisoryAI/Outputs/AdvisoryOutputStore.cs`), `FileSystemAdvisoryOutputStore` (`src/AdvisoryAi/StellaOps.AdvisoryAI.Hosting/FileSystemAdvisoryOutputStore.cs`)
## What's Missing
- **Playbook learning engine**: No dedicated module that analyzes past run outcomes to learn optimal remediation patterns and build reusable playbooks
- **Versioned playbook management**: No playbook versioning, publishing, or catalog system for sharing learned remediation workflows
- **Patch proposal generation**: No automated system that generates patch proposals by combining learned patterns from successful past remediations
- **Feedback loop learning**: No mechanism to feed PR merge/reject outcomes back into the learning engine to improve future recommendations
- **Playbook template library**: No library of reusable playbook templates (e.g., "upgrade-npm-dependency", "patch-container-base-image") with parameterization
## Implementation Plan
- Build a playbook learning engine that analyzes successful `Run` outcomes from `RunService`/`InMemoryRunStore`
- Add versioned playbook model with CRUD operations and a catalog API
- Implement patch proposal generation by matching current vulnerabilities against learned playbook patterns
- Add feedback loop from SCM connectors (PR merge/reject events) back to the learning engine
- Create a playbook template library with parameterized remediation workflows
## Related Documentation
- Source: Feature matrix scan