Claude Cursor GitHub Copilot opencode Skill

trust-compliance

Legal, security, privacy, and deliverability requirements every delivered app must satisfy — legal pages, PII/deletion rights, CSP/Trusted-Types/Zod, AI-agent security, RFC7807 errors, email domain auth.

LLM Mart · 0 points · 1 views 0 listing impressions 0 install-command copies
Download heymegabyte-agent-skills-27-trust-compliance-75f4661.zip · 1 KB

Install

skills CLI npx skills add https://github.com/heymegabyte/agent-skills/tree/main/27-trust-compliance
Claude Code claude plugin marketplace add https://llmmart.ai/marketplace.json && claude plugin install heymegabyte-agent-skills@llmmart
Git git clone https://github.com/heymegabyte/agent-skills.git

The skills CLI installs just this skill, for any of its supported agents. Claude Code installs the whole heymegabyte/agent-skills collection as a plugin from our marketplace. Git is the plain clone.

Skill manifest

Trust, Security & Compliance — every app must satisfy

Legal & policy

  • Ship real /privacy, /terms, /accessibility pages accurate to the ACTUAL data practices — never boilerplate that lies. Accessibility statement is ADA Title II / WCAG 2.2 AA-ready.
  • No dark patterns — honest defaults, no forced continuity, no confirm-shaming (Christ-like ethos).

Data & PII

  • Collect the minimum PII needed; never log secrets or PII in plaintext; keep credentials server-side only.
  • Honor user right-to-deletion and data export on request. State retention plainly.

Security controls (every app)

  • CSP Level 3 strict-dynamic + per-request nonce · Trusted Types · HSTS. No inline script without a nonce.
  • Zod validation at EVERY runtime boundary (env · API in/out · params · forms · webhooks · queue/DO messages · AI outputs) — validation is a security control, not just typing.
  • Forms gated by Turnstile. All hyperlinks valid (no dead/hostile links).
  • Tenant isolation: org_id on every row + query; 404 (never 403) on cross-tenant access.

AI-agent security

  • Treat all model/tool input as untrusted: defend against prompt injection; authorize every tool call; validate + schema-bind every AI output before use (Contract-First AI). Sandbox generated/risky code before it runs.

Errors (never leak internals)

  • User-facing errors use RFC7807 envelopes: code + correlationId + errors[] + what-to-do-next. Never expose stack traces, SQL, or internal identifiers.

Email domain authentication

  • Sending domain publishes SPF + DKIM + DMARC (BIMI where a VMC exists). These records live on the SENDING domain's zone (may differ from the site domain — surface any mismatch before staging).
Files (agent-skills)
  • SKILL.md 2.1 KB
    ---
    name: "trust-compliance"
    description: "Legal, security, privacy, and deliverability requirements every delivered app must satisfy — legal pages, PII/deletion rights, CSP/Trusted-Types/Zod, AI-agent security, RFC7807 errors, email domain auth."
    triggers:
      - "security"
      - "privacy"
      - "legal"
      - "compliance"
      - "csp"
    priority: 2
    pack: "security"
    stage: stable
    ---
    
    # Trust, Security & Compliance — every app must satisfy
    
    ## Legal & policy
    
    - Ship real `/privacy`, `/terms`, `/accessibility` pages accurate to the ACTUAL data practices — never boilerplate that lies. Accessibility statement is ADA Title II / WCAG 2.2 AA-ready.
    - No dark patterns — honest defaults, no forced continuity, no confirm-shaming (Christ-like ethos).
    
    ## Data & PII
    
    - Collect the minimum PII needed; never log secrets or PII in plaintext; keep credentials server-side only.
    - Honor user right-to-deletion and data export on request. State retention plainly.
    
    ## Security controls (every app)
    
    - CSP Level 3 `strict-dynamic` + per-request nonce · Trusted Types · HSTS. No inline script without a nonce.
    - Zod validation at EVERY runtime boundary (env · API in/out · params · forms · webhooks · queue/DO messages · AI outputs) — validation is a security control, not just typing.
    - Forms gated by Turnstile. All hyperlinks valid (no dead/hostile links).
    - Tenant isolation: `org_id` on every row + query; 404 (never 403) on cross-tenant access.
    
    ## AI-agent security
    
    - Treat all model/tool input as untrusted: defend against prompt injection; authorize every tool call; validate + schema-bind every AI output before use (Contract-First AI). Sandbox generated/risky code before it runs.
    
    ## Errors (never leak internals)
    
    - User-facing errors use RFC7807 envelopes: `code` + `correlationId` + `errors[]` + what-to-do-next. Never expose stack traces, SQL, or internal identifiers.
    
    ## Email domain authentication
    
    - Sending domain publishes SPF + DKIM + DMARC (BIMI where a VMC exists). These records live on the SENDING domain's zone (may differ from the site domain — surface any mismatch before staging).
    

Comments (0)

Sign in to join the conversation.

No comments yet.

Reviews (0)

No reviews yet.