Skill: Analyze Findings

A finding file bundles all of one rule's results. Read each result's code flow, split the bundle into distinct vulnerabilities, and give each a TP/FP verdict on its own evidence

Inputs

Provided by the caller, fall back to the default value when omitted. Ask back only when a required input is missing and has no sensible default

• project-root (optional) — root of the target project. Opentaint keeps all analysis artifacts under the fixed <project-root>/.opentaint/ directory, so every .opentaint/... path below resolves there. Default: current directory
• language (required) — target language for this project and language-specific instructions
• findings (required) — the finding file(s) to triage, each .opentaint/tracking/findings/<name>.yaml bundling one rule's SARIF results in sarif_hashes

Workflow

1. Reconcile before judging

A finding whose notes open with a reconcile line is a rescan result under a rule whose other findings are already triaged — most often the same vulnerability with a shifted hash, not a new one. Before judging it fresh, read the rule's already-triaged finding files and compare flows (source → sink, same essential path): if one matches, move this finding's sarif_hashes into that finding, drop this file, and let the inherited verdict stand — don't re-judge a flow already triaged. Only when no triaged finding matches do you treat it as new and continue below.

2. One result at a time — STOP checklist

For each hash in the bundle, before any verdict, read its raw result from .opentaint/results/report.sarif:

• find its SARIF result via sarif_hashes — each entry is the leading 16 chars of that result's vulnerabilitySourceSinkHash/vulnerabilityWithTraceHash fingerprint, so match it against the result's fingerprints/partialFingerprints — then read the raw codeFlows[]
• walk every step, source → hops → sink, confirming it's the same tainted value end to end; confirm the flow against the application source (the built project's own sources under .opentaint/project/sources/) and dependency code, not the trace text alone
• judge each result on its own trace — no verdict shared across results just because they share the rule

3. Split the bundle into logical findings

The results in the file all fired one rule, but may be several different vulnerabilities. Keep results that are the same vulnerability (same sink, same essential flow) together as one finding; move genuinely distinct ones into their own finding file with a new name and their sarif_hashes (per Tracking).

4. Classify and record

Verdict each logical finding from its flow:

• TP — the source is attacker-controlled, the sink is genuinely dangerous with that input, and nothing sanitizes it in between
• FP — a sanitizer/validator neutralizes it, the source isn't actually attacker-controlled (config, constant, server-set), the sink is safe for this input (parameterized, escaped), or the path is infeasible. Record which one

Set verdict and append the reasoning to notes, below the analyzer report already seeded there (per Tracking).

Output

Artifacts

• .opentaint/tracking/findings/<name>.yaml — each triaged finding with verdict set and the rationale appended to notes; a split also writes new finding file(s) (per Tracking)

Summary

• one line per finding: name, verdict, one-clause reason

Tracking

This skill writes only each finding's verdict and the reasoning appended to notes. A split additionally creates a new finding file — a fresh docker-like name, the moved sarifhashes, and ruleid copied from the bundle, carrying the seeded analyzer report into its notes and leaving poc pending. Never touch the poc field, or the sarif_hashes of a finding you keep.

.opentaint/tracking/findings/<name>.yaml — one finding, bundling one rule's SARIF results and carrying it through triage and PoC. A script seeds each file from the scan — its sarifhashes, ruleid, and the analyzer report in notes. Triage sets verdict and appends its reasoning. The PoC stage sets poc and appends its outcome. Keep it clear from comments

sarif_hashes: [a1b2c3d4, e5f6a7b8]
rule_id: java/security/sqli.yaml:sqli
verdict: TP
notes: >
  <analyzer report for these results — seeded from the scan>
  triage: @RequestParam orderBy is attacker-controlled; reaches ${} in SelectProvider unsanitized → TP
  poc: logged in as a seeded user, then GET /api/orders?orderBy=id);SELECT pg_sleep(5)-- delayed ~5s → confirmed
poc: confirmed

Constraints

• Verdicts and notes go in the finding files only — never write .opentaint/vulnerabilities.md; the orchestrator assembles it from the verdicts
• Judge each result on its own trace, never share one verdict across results just because they fired the same rule

Gotchas

• Bulk verdicts are the most common triage error — many results marked under one shared rationale with the traces unread
• A rule's bundle is not one finding — split distinct vulnerabilities apart, but keep true duplicates (same sink and flow) together as one finding with multiple sarif_hashes