How ppd-ibm/vulnerability-agentic-harness works
VVAH is an agentic vulnerability-discovery, remediation, and fix-validation harness. Its principal workflow scans a target repository through discovery, threat modeling, analysis, verification, and finding synthesis; remediation can then propose or apply fixes, followed by adversarial validation. The README documents all eleven stages and Markdown/SARIF outputs. Sampled code confirms the scan orchestrator’s stage wiring, model-backend dispatch, remediation plugin invocation, and validation execution. Some stage internals are unsampled, so their documented sequence is shown without asserting unobserved internal calls.
Main entry points: scan.py Scan orchestrator · cli.py Command interface · runner.py Remediation runner · pipeline.py Validation execution
Components 29
Discovery and Scan
- Scan orchestrator
vvaharness/orchestrator/scan.py - Preprocess and scope
vvaharness/pipeline/stages/s1_preprocess.py - Threat modeling
vvaharness/pipeline/stages - Hunting plan
vvaharness/pipeline/stages/s3_decompose.py - Deep analysis
vvaharness/pipeline/stages/s4_deepdive.py - Adversarial verification
vvaharness/pipeline/stages/s6_verify.py - Dedup and chain
vvaharness/pipeline/stages - Finding reports
vvaharness/models.py
Models and Integrations
- Finding enrichment
vvaharness/report/enrich.py - Sensitive-data redaction
vvaharness/report/redact.py - Language hints
vvaharness/lang/hints.py - Finding data models
vvaharness/models.py - Model backend routing
vvaharness/backends/llm.py - Run checkpoints
vvaharness/orchestrator/checkpoints.py - Configuration paths
vvaharness/orchestrator/config_paths.py - Run state store
vvaharness/orchestrator/store.py
Remediation
- Remediation runner
vvaharness/remediation_agent/runner.py - Finding parser
vvaharness/remediation_agent/report_parser/parse.py - Patch plugin
vvaharness/remediation_agent/plugin_runner/run.py - Patch artifacts
vvaharness/remediation_agent/artifacts
Validation
- Validation command
vvaharness/validation/cli/_run.py - Validation execution
vvaharness/validation/execution/pipeline.py - Agent validation session
vvaharness/validation/session/launcher.py - Result collection
vvaharness/validation/io/result_collector.py - Fix-quality scoring
vvaharness/validation/scoring/_engine.py - Validation model clients
vvaharness/validation/backends
Other components
- Authorized engineer
- Target repository
- Command interface
vvaharness/cli.py
Connections 34
- Authorized engineer → Command interface: runs commands
README.md - Command interface → Scan orchestrator: starts scan
vvaharness/cli.py - Scan orchestrator → Target repository: scans code
README.md - Scan orchestrator → Preprocess and scope: runs stage
vvaharness/orchestrator/scan.py - Scan orchestrator → Threat modeling: runs stage
vvaharness/orchestrator/scan.py - Scan orchestrator → Hunting plan: runs stage
vvaharness/orchestrator/scan.py - Scan orchestrator → Deep analysis: runs stage
vvaharness/orchestrator/scan.py - Scan orchestrator → Adversarial verification: runs stage
vvaharness/orchestrator/scan.py - Scan orchestrator → Dedup and chain: runs stages
vvaharness/orchestrator/scan.py - Preprocess and scope → Threat modeling: precedes
README.md - Threat modeling → Hunting plan: precedes
README.md - Hunting plan → Deep analysis: precedes
README.md - Deep analysis → Adversarial verification: precedes
README.md - Adversarial verification → Dedup and chain: precedes
README.md - Scan orchestrator → Model backend routing: dispatches models
vvaharness/orchestrator/scan.py - Model backend routing → Finding data models: uses shared types
- Dedup and chain → Finding reports: emits findings
README.md - Scan orchestrator → Finding enrichment: enriches findings
vvaharness/orchestrator/scan.py - Scan orchestrator → Run checkpoints: uses checkpoints
vvaharness/orchestrator/scan.py - Command interface → Configuration paths: resolves configuration
vvaharness/cli.py - Command interface → Run state store: uses run state
vvaharness/cli.py - Command interface → Remediation runner: starts remediation
vvaharness/cli.py - Remediation runner → Finding parser: parses findings
vvaharness/remediation_agent/runner.py - Remediation runner → Patch plugin: invokes plugin
vvaharness/remediation_agent/runner.py - Patch plugin → Patch artifacts: writes patch artifacts
vvaharness/remediation_agent/plugin_runner/run.py - Remediation runner → Run checkpoints: saves progress
vvaharness/remediation_agent/runner.py - Command interface → Validation command: starts validation
README.md - Validation command → Validation execution: executes plans
- Validation execution → Agent validation session: launches session
vvaharness/validation/execution/pipeline.py - Validation execution → Result collection: collects results
vvaharness/validation/execution/pipeline.py - Validation execution → Fix-quality scoring: returns validation results
- Agent validation session → Validation model clients: uses model client
- Finding reports → Remediation runner: provides findings
README.md - Patch artifacts → Validation command: supplies remediation DTOs
README.md