v0.1.5

Added

• Added fixtures for direct provider error bodies, generic HTTP Retry-After
  handling, OpenAI billing hard-limit errors, Anthropic billing errors and
  Gemini quota/billing exhaustion.
• Added fixtures for mixed billing/rate-limit signals, generic non-retryable
  Retry-After headers and Gemini rate-bucket quota exhaustion.

Changed

• Preserve Retry-After / retry-after-ms delays for provider-unknown HTTP
  errors.
• Detect direct provider error bodies, case-insensitive Map headers and common
  Node-style numeric or array header values.
• Classify more provider code/type strings without requiring an HTTP status,
  including quota, permission, not-found, timeout, overload and server failures.
• Only expose provider retry delays on retryable normalized errors, validate
  HTTP status candidates and reject invalid retry delay values.

Published to npm as llm-errors@0.1.5.

llm-errors v0.1.4

Added

• Public provider error fixture corpus for OpenAI, Anthropic, Gemini and transport failures.
• Fixture-driven regression tests pairing SDK-like/fetch-like inputs with expected normalized outputs.
• Published fixtures/ in the npm package.

Verification

• PR: #9
• npm: https://www.npmjs.com/package/llm-errors/v/0.1.4
• Closes: #1

llm-errors v0.1.3

Changed

• Updated vulnerable development tooling to clear Dependabot alerts.
• Added CodeQL, OpenSSF Scorecard, pinned GitHub Actions, least-privilege workflow permissions, Dependabot config and a Scorecard README badge.

Verification

• CI, CodeQL, Scorecard and Release workflows passed.
• npm: https://www.npmjs.com/package/llm-errors/v/0.1.3

llm-errors v0.1.2

Docs/package metadata release.

• Publishes README package-status badges and npm download visibility to the npm package page.
• Adds release notes for the current patch version.
• No runtime behavior changes.

llm-errors v0.1.1

Latest npm release for llm-errors.

Highlights:

• Normalizes OpenAI, Anthropic and Gemini API errors into one portable shape.
• Exposes category, retryable status and retry delay helpers for provider-agnostic retry loops.
• Keeps the package zero-dependency and MIT licensed.
• Adds OpenAI-compatible provider keywords for discovery across agent/tooling ecosystems.