Skip to content

Latest commit

 

History

History
36 lines (27 loc) · 1.8 KB

File metadata and controls

36 lines (27 loc) · 1.8 KB

Failure Model

Purpose: turn predictable failures into defenses, validators, or required evidence.

Failure Inventory

Failure Why likely Cost if missed Defense Validator or evidence Status
SDK drift Agent invents or changes SDK shape App fails in ChatGPT Approved recipes and SDK API validator sdk-api-check Pending
Missing intent Destination unclear Wrong compliance standard Deployment Intent gate profile-check Pending
Unsafe tool Tool mutates more than expected Data loss or wrong user action Tool contract and hints tool-contract-check Pending
Widget drift UI state diverges from MCP output Broken UX structuredContent and widget state checks widget-check Pending
Secret exposure Credentials leak into client or repo Account compromise No secrets in code and secret scan security-check Pending
PII leakage Logs or structuredContent expose data Privacy failure Data minimization and redaction evidence-check Pending
Public review mismatch App claims differ from behavior Rejection or user harm Golden and negative prompts golden-prompts.md Pending
Weak app value App does not add know/do/show value App feels useless in ChatGPT App value contract and focused capability surface app-value-check Pending

Asymmetric Defenses

  • Contracts before code.
  • Profile-specific compliance.
  • Know/do/show value gate.
  • Static validators before handoff.
  • Golden and negative prompts.
  • Verification report tied to Deployment Intent.

Open Weaknesses

  • Live ChatGPT connector status: Not verified.
  • Cloudflare Tunnel status: Not verified.
  • Production approval status: Not approved.

Decision

  • Status: FAIL
  • Reason: Failure model has not been updated for the final app.