v0.2.1

Fixed

• Preserve a user-selected sample while the WASM runtime initializes and keep
  analysis disabled until the runtime is ready.

Security

• Redact credentials inside JSON-encoded output strings and keep parser or
  redaction-boundary failures from echoing user values.
• Redact free-text authorization headers and sensitive key: value or
  key=value assignments while preserving benign investigation context.

v0.2.0

Added

• Timestamped OpenAI Agents SDK, LangChain, and PydanticAI/Logfire importers
  with structural auto-detection and pinned upstream contract provenance.
• A local tool-call-trace check CLI for files and stdin, explicit formats,
  normalized JSON output, redaction, and stable failure codes.
• Browser format selection and opt-in redaction with exact JSON Pointer paths,
  replacement counts, and reparseable redacted Generic JSON.

Security

• Added deterministic redaction for authorization, X-API-Key, token,
  password, secret, and private-key-like fields plus sensitive HTTP(S) URL
  components.
• Added a 100,000-line input limit and secret-free CLI and browser failures
  whenever redaction is requested.
• Added adversarial Rust, WASM, CLI, and four-viewport browser coverage for
  redaction boundaries and preserved searchable IDs.

v0.1.0

Added

• Strict Generic JSON and timestamped OpenAI run-step parsing.
• Status-specific OpenAI terminal timestamps and preserved inline function
  output.
• Browser-local WASM timeline, call details, statistics, duplicate detection,
  and slow-call findings.
• Real WASM boundary coverage and four-viewport Chromium tests.
• Reusable Rust, WASM, and supply-chain CI workflows.

Security

• Added a 5 MiB input limit, a 2,000-call limit, unique-ID validation, checked
  timestamp conversion, overflow-safe analysis, text-only DOM rendering, and
  a no-external-request browser gate.

Removed

• Removed fabricated Anthropic timing support and unimplemented MCP tool
  declarations.