Risk Scoring

Every AI action in GitWire is scored before execution. Low-confidence and high-risk actions are blocked or downgraded automatically.

How It Works

graph TD
    A[AI produces diagnosis/fix] --> B{Confidence >= threshold?}
    B -->|No| C["Comment-only (no mutation)"]
    B -->|Yes| D{Risk score?}
    D -->|High >=50| E["Block + explain"]
    D -->|Medium 25-49| F["Execute with warning"]
    D -->|Low <25| G["Execute normally"]

CI Healing Risk Scoring

When a CI run fails and Claude diagnoses it, the diagnosis is scored:

Factor	Points	When Applied
Low AI confidence	+40	confidence: "low"
Medium AI confidence	+15	confidence: "medium"
High-risk failure type	+30	infra_error, build_error, unknown
Other failure type	+15	test_flaky, dependency_missing, etc.
No failing file identified	+20	failing_file: null
Not auto-fixable	+30	auto_fixable: false

Risk Levels

Score	Level	Action
0–24	Low	Patch PR created normally
25–49	Medium	Patch PR created, warning logged
50–100	High	Blocked — only diagnosis comment posted

Confidence Gate

Before risk scoring, the confidence gate checks:

pillars:
  ci_healing:
    min_confidence_to_patch: medium  # low | medium | high

If Claude's confidence is below the threshold, no patch is attempted. Instead, a comment-only diagnosis is posted to the commit/PR:

🤖 GitWire CI Diagnosis (confidence: low, threshold: medium)

Confidence below threshold — posting diagnosis only. Adjust ci_healing.min_confidence_to_patch in .gitwire.yml if needed.

Issue Fix Risk Scoring

When the autonomous contributor analyzes an issue and produces a fix:

Factor	Points	When Applied
Complex issue	+40	complexity: "complex"
Moderate complexity	+20	complexity: "moderate"
>3 files changed	+20	Multi-file fix
>30% line delta	+15	Large changes relative to file size

Pre-Check

Complexity is mapped to confidence:

Complexity	Confidence
trivial	high
simple	medium
moderate / complex	low

If this confidence is below min_confidence_to_submit, the fix is rejected before PR creation.

High-Risk Block

Fixes scoring ≥ 50 are blocked with a detailed rejection:

🚫 GitWire Fix — high risk

Risk score: 65/100

• Complex issue
• 4 files changed
• src/app.js: 45% lines changed

This fix is too risky for autonomous submission.

Configuration

.gitwire.yml

pillars:
  ci_healing:
    min_confidence_to_patch: high    # Only high-confidence patches
  issue_fix:
    min_confidence_to_submit: low    # Allow low-confidence attempts

Dashboard UI

Navigate to Config → select a repo → find the CI Healing or Autonomous Contributor card → adjust Min confidence (1-3):

Value	Meaning
1 (low)	All AI actions allowed
2 (medium)	Block low-confidence (default)
3 (high)	Only high-confidence actions

API

PATCH /api/config/:owner/:repo
{
  "pillars": {
    "ci_healing": { "min_confidence_to_patch": "high" },
    "issue_fix": { "min_confidence_to_submit": "low" }
  }
}

Audit Trail

All risk assessments are logged:

{
  "level": 30,
  "msg": "CI risk assessment",
  "runId": 12345,
  "riskScore": 55,
  "riskLevel": "high",
  "minConfidence": "medium"
}

{
  "level": 30,
  "msg": "Fix risk assessment",
  "repo": "org/repo",
  "issueNumber": 42,
  "riskScore": 65,
  "riskLevel": "high",
  "reasons": ["Complex issue", "4 files changed"]
}

→ Policy-as-Code | Dry Run Mode | CI Healing