Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Apply security best practices for Fondo including OAuth token management, financial data protection, SOC 2 compliance, and access control. Trigger: "fondo security", "fondo data protection", "fondo SOC 2", "fondo access control".
.claude/skills/jeremylongshore-fondo-security-basics/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 17% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -6% | 0% |
Fondo handles startup tax preparation, bookkeeping, and R&D tax credits containing SSNs, EINs, bank account details, revenue figures, and complete tax returns. A breach exposes founder personal tax data, company financials, and IRS filing details. Protect OAuth connections to banking/payroll systems, exported financial documents, and team access controls with the same rigor as a CPA firm.
typescriptfunction createFondoClient(): { apiKey: string; baseUrl: string } { const apiKey = process.env.FONDO_API_KEY; if (!apiKey) { throw new Error("Missing FONDO_API_KEY — store in secrets manager, never in code"); } // Fondo keys access tax returns and SSN/EIN data — treat as highest sensitivity console.log("Fondo client initialized (key suffix:", apiKey.slice(-4), ")"); return { apiKey, baseUrl: "https://api.fondo.com/v1" }; }
typescriptimport crypto from "crypto"; import { Request, Response, NextFunction } from "express"; function verifyFondoWebhook(req: Request, res: Response, next: NextFunction): void { const signature = req.headers["x-fondo-signature"] as string; const secret = process.env.FONDO_WEBHOOK_SECRET!; const expected = crypto.createHmac("sha256", secret).update(req.body).digest("hex"); if (!signature || !crypto.timingSafeEqual(Buffer.from(signature), Buffer.from(expected))) { res.status(401).send("Invalid signature"); return; } next(); }
typescriptimport { z } from "zod"; const TaxFilingSchema = z.object({ entity_id: z.string().uuid(), tax_year: z.number().int().min(2015).max(2030), filing_type: z.enum(["1120", "1120S", "1065", "941", "R&D_credit"]), ein: z.string().regex(/^\d{2}-\d{7}$/), revenue: z.number().nonnegative(), status: z.enum(["draft", "review", "filed", "amended"]), }); function validateTaxFiling(data: unknown) { return TaxFilingSchema.parse(data); }
typescriptconst FONDO_PII_FIELDS = ["ssn", "ein", "bank_account", "routing_number", "tax_return_url", "revenue", "salary"]; function redactFondoLog(record: Record<string, unknown>): Record<string, unknown> { const redacted = { ...record }; for (const field of FONDO_PII_FIELDS) { if (field in redacted) redacted[field] = "[REDACTED]"; } return redacted; }
.gitignore enforced)| Vulnerability | Risk | Mitigation | |---|---|---| | Leaked API key | Full access to tax returns and SSN/EIN data | Secrets manager + rotation | | Unencrypted financial exports | Tax data exposed via CSV on disk | GPG encryption + secure deletion | | Stale OAuth tokens | Compromised banking/payroll connections | Quarterly OAuth review + revocation | | Overly broad team access | Viewer sees SSN/salary data | Role-based access control enforcement | | Tax data in application logs | IRS compliance violation | Field-level PII redaction |
See fondo-prod-checklist.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 19,579 | 17,370 | -11% | 1 | 1 | 0% | 2,912 | 3,572 | +23% | 0 | 0 | — |
case-02 | fail→fail | 18,111 | 19,831 | +9% | 1 | 1 | 0% | 3,773 | 4,118 | +9% | 0 | 0 | — |
case-03 | fail→pass | 18,348 | 14,771 | -19% | 1 | 1 | 0% | 3,558 | 3,998 | +12% | 0 | 0 | — |
case-04 | fail→pass | 17,533 | 8,595 | -51% | 1 | 1 | 0% | 2,236 | 2,609 | +17% | 0 | 0 | — |
case-05 | pass→pass | 13,216 | 6,915 | -48% | 1 | 1 | 0% | 2,149 | 2,168 | +1% | 0 | 0 | — |
case-06 | fail→pass | 16,697 | 3,741 | -78% | 1 | 1 | 0% | 2,053 | 1,662 | -19% | 0 | 0 | — |
case-07 | pass→fail | 19,237 | 15,101 | -22% | 1 | 1 | 0% | 2,399 | 3,598 | +50% | 0 | 0 | — |
case-08 | pass→pass | 19,078 | 10,948 | -43% | 1 | 1 | 0% | 2,143 | 2,678 | +25% | 0 | 0 | — |
case-09 | pass→pass | 20,904 | 19,131 | -8% | 1 | 1 | 0% | 2,586 | 3,530 | +37% | 0 | 0 | — |
case-10 | pass→pass | 15,148 | 10,580 | -30% | 1 | 1 | 0% | 1,715 | 2,722 | +59% | 0 | 0 | — |
case-11 | fail→fail | 18,490 | 22,210 | +20% | 1 | 1 | 0% | 2,045 | 3,762 | +84% | 0 | 0 | — |
case-12 | pass→pass | 19,635 | 21,761 | +11% | 1 | 1 | 0% | 2,418 | 4,468 | +85% | 0 | 0 | — |
case-13 | fail→pass | 10,886 | 9,289 | -15% | 1 | 1 | 0% | 1,864 | 1,753 | -6% | 0 | 0 | — |
case-14 | pass→pass | 10,719 | 8,317 | -22% | 1 | 1 | 0% | 2,042 | 2,514 | +23% | 0 | 0 | — |
case-15 | fail→pass | 15,129 | 18,110 | +20% | 1 | 1 | 0% | 2,571 | 3,453 | +34% | 0 | 0 | — |
case-16 | fail→pass | 16,053 | 6,089 | -62% | 1 | 1 | 0% | 2,426 | 2,185 | -10% | 0 | 0 | — |
case-17 | fail→pass | 10,115 | 13,148 | +30% | 1 | 1 | 0% | 2,033 | 2,623 | +29% | 0 | 0 | — |
case-18 | fail→fail | 18,405 | 17,587 | -4% | 1 | 1 | 0% | 2,551 | 3,429 | +34% | 0 | 0 | — |
case-19 | pass→fail | 16,751 | 10,958 | -35% | 1 | 1 | 0% | 2,190 | 3,136 | +43% | 0 | 0 | — |
case-20 | pass→pass | 11,332 | 9,292 | -18% | 1 | 1 | 0% | 2,157 | 2,908 | +35% | 0 | 0 | — |
case-21 | fail→pass | 9,332 | 1,712 | -82% | 1 | 1 | 0% | 933 | 1,429 | +53% | 0 | 0 | — |
case-22 | pass→pass | 8,833 | 9,292 | +5% | 1 | 1 | 0% | 2,064 | 3,281 | +59% | 0 | 0 | — |
case-23 | pass→pass | 18,873 | 16,168 | -14% | 1 | 1 | 0% | 3,095 | 4,754 | +54% | 0 | 0 | — |
case-24 | pass→pass | 17,879 | 10,305 | -42% | 1 | 1 | 0% | 2,414 | 3,305 | +37% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted. The headline lift of +29 percentage points is the difference between those two pass rates over the 24 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.