FFlowRail
Sign inGet started
Docs/Skills & MCP tools

flowrail-triage

v1 roadmap

On the v1 roadmap, not in the POC you install today. Triages a scanner finding against the design review and the codebase. Returns a verdict (true positive, false positive, accepted risk) with a confidence score and an evidence digest. Detects guardrail escapes and feeds them to the self-improvement loop.

01

When it runs

Invoked by the scanner orchestrator on every finding ingested, or directly when a developer asks Claude to triage a specific scanner output. Runs customer-side with codebase access; only the verdict and evidence digest leave your environment.

What you see in Claude Codev0.4.2
↳ flowrail-triage  fnd_d4a1c8e2 (semgrep · javascript.express.injection.tainted-sql-string)
  verdict       false_positive  (confidence 0.87)
  evidence
    user_input sanitized at src/api/orders.ts:38 via zod
    flagged sink at line 47 receives only validated values
    no taint flow from request body to query parameter
  guardrail_escape  no
MCP toolv0.4.2
flowrail_triage_finding
Requestv0.4.2
{
  "design_review_id": "dr_7f3a9e2c",
  "finding": {
    "finding_id":  "fnd_d4a1c8e2",
    "scanner":     "semgrep",
    "rule_id":     "javascript.express.injection.tainted-sql-string",
    "file":        "src/api/orders.ts",
    "line_range":  [42, 51],
    "code_hash":   "sha256:e3b0c44…"
  }
}
Responsev0.4.2
{
  "verdict":         "false_positive",
  "confidence":      0.87,
  "evidence":        ["...", "..."],
  "guardrail_escape": false,
  "proposed_guardrail_update": null
}

When a finding is a true positive AND its threat was predicted at design review AND no guardrail blocked the write, the response carries proposed_guardrail_update with a candidate tightening. The proposal goes through the integrity gate (rate limit, FPR replay, attribution) before any auto-merge.