feat: add Oathe behavioral security scan integration

convex/_generated/api.d.ts

@@ -76,6 +76,7 @@ import type * as lib_userSearch from "../lib/userSearch.js";
 import type * as lib_webhooks from "../lib/webhooks.js";
 import type * as llmEval from "../llmEval.js";
 import type * as maintenance from "../maintenance.js";
+import type * as oathe from "../oathe.js";
 import type * as rateLimits from "../rateLimits.js";
 import type * as search from "../search.js";
 import type * as seed from "../seed.js";
@@ -170,6 +171,7 @@ declare const fullApi: ApiFromModules<{
   "lib/webhooks": typeof lib_webhooks;
   llmEval: typeof llmEval;
   maintenance: typeof maintenance;
+  oathe: typeof oathe;
   rateLimits: typeof rateLimits;
   search: typeof search;
   seed: typeof seed;

convex/crons.ts

@@ -59,6 +59,13 @@ crons.interval('vt-cache-backfill', { minutes: 30 }, internal.vt.backfillActiveS
 // Daily re-scan of all active skills at 3am UTC
 crons.daily('vt-daily-rescan', { hourUTC: 3, minuteUTC: 0 }, internal.vt.rescanActiveSkills, {})
 
+crons.interval(
+  'oathe-pending-results',
+  { minutes: 10 },
+  internal.oathe.fetchPendingOatheResults,
+  { batchSize: 50 },
+)
+
 crons.interval(
   'download-dedupe-prune',
   { hours: 24 },

convex/lib/globalStats.ts

@@ -91,6 +91,10 @@ export async function setGlobalPublicSkillsCount(
   }
 }
 
+/**
+ * IMPORTANT: Must be called AFTER ctx.db.patch() — the fallback recount
+ * reads post-patch DB state. Calling before patch produces wrong counts.
+ */
 export async function adjustGlobalPublicSkillsCount(
   ctx: GlobalStatsWriteCtx,
   delta: number,
@@ -106,13 +110,19 @@ export async function adjustGlobalPublicSkillsCount(
       }
     | null
     | undefined
+  // NOTE: All visibility mutations read/write this single row. Under high concurrent
+  // writes, Convex OCC retries increase. Acceptable at current scale; if contention
+  // becomes an issue, consider sharding by key prefix or batching deltas.
   try {
     existing = await ctx.db
       .query('globalStats')
       .withIndex('by_key', (q) => q.eq('key', GLOBAL_STATS_KEY))
       .unique()
   } catch (error) {
-    if (isGlobalStatsStorageNotReadyError(error)) return
+    if (isGlobalStatsStorageNotReadyError(error)) {
+      console.warn('[globalStats] Storage not ready — delta adjustment skipped:', normalizedDelta)
+      return
+    }
     throw error
   }
 

convex/lib/skillPublish.ts

@@ -291,6 +291,10 @@ export async function publishVersionForUser(
     versionId: publishResult.versionId,
   })
 
+  await ctx.scheduler.runAfter(0, internal.oathe.notifyOathe, {
+    versionId: publishResult.versionId,
+  })
+
   const ownerHandle = owner?.handle ?? owner?.displayName ?? owner?.name ?? 'unknown'
 
   if (!options.skipBackup) {

convex/oathe.test.ts

@@ -0,0 +1,181 @@
+/* @vitest-environment node */
+import { describe, expect, it } from 'vitest'
+import { __test, mapReportToAnalysis } from './oathe'
+
+const { scoreToRating, verdictToStatus, DIMENSION_LABELS } = __test
+
+describe('scoreToRating', () => {
+  it('returns ok for scores >= 80', () => {
+    expect(scoreToRating(80)).toBe('ok')
+    expect(scoreToRating(100)).toBe('ok')
+    expect(scoreToRating(95)).toBe('ok')
+  })
+
+  it('returns note for scores 50–79', () => {
+    expect(scoreToRating(50)).toBe('note')
+    expect(scoreToRating(79)).toBe('note')
+    expect(scoreToRating(65)).toBe('note')
+  })
+
+  it('returns concern for scores 20–49', () => {
+    expect(scoreToRating(20)).toBe('concern')
+    expect(scoreToRating(49)).toBe('concern')
+    expect(scoreToRating(35)).toBe('concern')
+  })
+
+  it('returns danger for scores < 20', () => {
+    expect(scoreToRating(0)).toBe('danger')
+    expect(scoreToRating(19)).toBe('danger')
+    expect(scoreToRating(10)).toBe('danger')
+  })
+
+  it('handles boundary values exactly', () => {
+    expect(scoreToRating(80)).toBe('ok')
+    expect(scoreToRating(79)).toBe('note')
+    expect(scoreToRating(50)).toBe('note')
+    expect(scoreToRating(49)).toBe('concern')
+    expect(scoreToRating(20)).toBe('concern')
+    expect(scoreToRating(19)).toBe('danger')
+  })
+})
+
+describe('verdictToStatus', () => {
+  it('maps SAFE verdict', () => {
+    expect(verdictToStatus('SAFE')).toBe('safe')
+    expect(verdictToStatus('safe')).toBe('safe')
+    expect(verdictToStatus('Safe')).toBe('safe')
+  })
+
+  it('maps CAUTION verdict', () => {
+    expect(verdictToStatus('CAUTION')).toBe('caution')
+    expect(verdictToStatus('caution')).toBe('caution')
+  })
+
+  it('maps DANGEROUS verdict', () => {
+    expect(verdictToStatus('DANGEROUS')).toBe('dangerous')
+    expect(verdictToStatus('dangerous')).toBe('dangerous')
+  })
+
+  it('maps MALICIOUS verdict', () => {
+    expect(verdictToStatus('MALICIOUS')).toBe('malicious')
+    expect(verdictToStatus('malicious')).toBe('malicious')
+  })
+
+  it('returns unknown for unrecognized verdicts', () => {
+    expect(verdictToStatus('UNKNOWN')).toBe('unknown')
+    expect(verdictToStatus('')).toBe('unknown')
+    expect(verdictToStatus('something-else')).toBe('unknown')
+  })
+})
+
+describe('mapReportToAnalysis', () => {
+  const baseReport = {
+    audit_id: 'audit-123',
+    skill_url: 'https://clawhub.ai/test-owner/test-skill',
+    skill_slug: 'test-skill',
+    summary: 'No significant threats detected.',
+    recommendation: 'Safe to use.',
+    trust_score: 92,
+    verdict: 'SAFE',
+    category_scores: {
+      prompt_injection: {
+        score: 95,
+        weight: 1,
+        findings: [],
+      },
+      data_exfiltration: {
+        score: 88,
+        weight: 1,
+        findings: ['Minor outbound request detected'],
+      },
+    },
+    findings: [],
+  }
+
+  it('maps a complete report to analysis object', () => {
+    const result = mapReportToAnalysis(baseReport, 'test-owner/test-skill')
+
+    expect(result.status).toBe('safe')
+    expect(result.score).toBe(92)
+    expect(result.verdict).toBe('SAFE')
+    expect(result.summary).toBe('No significant threats detected.')
+    expect(result.reportUrl).toBe('https://oathe.ai/report/test-owner/test-skill')
+    expect(result.checkedAt).toBeGreaterThan(0)
+  })
+
+  it('maps dimensions with correct labels and ratings', () => {
+    const result = mapReportToAnalysis(baseReport, 'test-owner/test-skill')
+
+    expect(result.dimensions).toHaveLength(2)
+
+    const piDim = result.dimensions.find((d) => d.name === 'prompt_injection')
+    expect(piDim).toBeDefined()
+    expect(piDim!.label).toBe('Prompt Injection')
+    expect(piDim!.rating).toBe('ok')
+    expect(piDim!.detail).toBe('No issues detected. Score: 95/100')
+
+    const deDim = result.dimensions.find((d) => d.name === 'data_exfiltration')
+    expect(deDim).toBeDefined()
+    expect(deDim!.label).toBe('Data Exfiltration')
+    expect(deDim!.rating).toBe('ok')
+    expect(deDim!.detail).toBe('Minor outbound request detected')
+  })
+
+  it('uses dimension key as label fallback for unknown dimensions', () => {
+    const report = {
+      ...baseReport,
+      category_scores: {
+        custom_dimension: { score: 60, weight: 1, findings: [] },
+      },
+    }
+    const result = mapReportToAnalysis(report, 'test-owner/test-skill')
+
+    const dim = result.dimensions.find((d) => d.name === 'custom_dimension')
+    expect(dim!.label).toBe('custom_dimension')
+  })
+
+  it('maps CAUTION verdict correctly', () => {
+    const report = { ...baseReport, verdict: 'CAUTION', trust_score: 54 }
+    const result = mapReportToAnalysis(report, 'test-owner/test-skill')
+
+    expect(result.status).toBe('caution')
+    expect(result.score).toBe(54)
+  })
+
+  it('maps MALICIOUS verdict correctly', () => {
+    const report = { ...baseReport, verdict: 'MALICIOUS', trust_score: 12 }
+    const result = mapReportToAnalysis(report, 'test-owner/test-skill')
+
+    expect(result.status).toBe('malicious')
+    expect(result.score).toBe(12)
+  })
+
+  it('uses first finding as detail when findings exist', () => {
+    const report = {
+      ...baseReport,
+      category_scores: {
+        code_execution: {
+          score: 30,
+          weight: 1,
+          findings: ['Subprocess spawned', 'File written to /tmp'],
+        },
+      },
+    }
+    const result = mapReportToAnalysis(report, 'test-owner/test-skill')
+
+    const dim = result.dimensions.find((d) => d.name === 'code_execution')
+    expect(dim!.detail).toBe('Subprocess spawned')
+    expect(dim!.rating).toBe('concern')
+  })
+})
+
+describe('DIMENSION_LABELS', () => {
+  it('has labels for all standard dimensions', () => {
+    expect(DIMENSION_LABELS.prompt_injection).toBe('Prompt Injection')
+    expect(DIMENSION_LABELS.data_exfiltration).toBe('Data Exfiltration')
+    expect(DIMENSION_LABELS.code_execution).toBe('Code Execution')
+    expect(DIMENSION_LABELS.clone_behavior).toBe('Clone Behavior')
+    expect(DIMENSION_LABELS.canary_integrity).toBe('Canary Integrity')
+    expect(DIMENSION_LABELS.behavioral_reasoning).toBe('Behavioral Reasoning')
+  })
+})