v2.1.0 — @debriefer/browser auth & sessions

v2.1.0

@debriefer/browser — Browser auth, sessions, and login handlers (NEW features)

Phase 2 additions:

• Config loader: Environment variable-based credentials and CAPTCHA solver configuration
• Session manager: Cookie persistence to disk with TTL, 0o600 file permissions, 0o700 directory permissions
• Base login handler: Abstract class with CAPTCHA detection/solving and session verification
• NYTimes login handler: Two-step email→password flow with DataDome CAPTCHA handling
• Washington Post login handler: Two-step login with flexible selectors
• createBrowserDefaults() factory: Combines fetchPage + login + session management

Phase 1 (from v2.0.2):

• Browser stealth with fingerprint-injector
• CAPTCHA detection (reCAPTCHA v2/v3, hCaptcha, PerimeterX, DataDome)
• CAPTCHA solving (2Captcha, CapSolver)
• Archive fallbacks (archive.org, archive.is)
• createBrowserFetchPage() factory

Usage

import { createBrowserDefaults } from "@debriefer/browser"

const browser = createBrowserDefaults({
  captchaSolver: { provider: "2captcha", apiKey: "..." },
  credentials: { "nytimes.com": { email: "...", password: "..." } },
})

const google = googleSearch({ fetchPage: browser.fetchPage })
await browser.login("nytimes.com")

v2.0.2 — @debriefer/browser

v2.0.2

Publishes @debriefer/browser@2.0.0 — browser stealth, CAPTCHA solving, and archive fallbacks.

@debriefer/browser (NEW)

• Browser stealth: Playwright + fingerprint-injector for realistic fingerprints
• CAPTCHA detection: reCAPTCHA v2/v3, hCaptcha, PerimeterX, DataDome
• CAPTCHA solving: 2Captcha and CapSolver integration
• Archive fallbacks: archive.org + archive.is with rate limiting
• 4-step fetch chain: direct → archive.org → archive.is → browser + CAPTCHA
• createBrowserFetchPage() factory compatible with WebSearchOptions.fetchPage

@debriefer/sources

• Removed playwright-core and fingerprint-injector from optional deps (moved to @debriefer/browser)

v2.0.1 — publish cli, server, mcp

v2.0.1

Publishes the remaining three packages to npm:

• @debriefer/cli@2.0.0 — CLI tool (npx @debriefer/cli debrief "Marie Curie")
• @debriefer/server@2.0.0 — Express HTTP server
• @debriefer/mcp@2.0.0 — MCP server for AI assistants

No code changes from v2.0.0 — this release adds cli/server/mcp to the publish workflow.

v2.0.0 — AI-first defaults, @debriefer scope

What's New

@debriefer/ai (1.0.0) — NEW

• createAIDefaults() factory returning pre-built Claude Haiku callbacks for section filtering, confidence scoring, link selection, and person validation
• ClaudeSynthesizer (moved from core) — AI synthesis via Anthropic SDK
• AIClient interface + HaikuClient default for pluggable AI providers
• Graceful fallback to heuristics when ANTHROPIC_API_KEY is missing or AI calls fail

@debriefer/core (2.0.0) — BREAKING

• Breaking: ClaudeSynthesizer moved to @debriefer/ai — update imports
• Breaking: @anthropic-ai/sdk peer dependency removed — core is now fully zero-AI
• Added confidenceScorer callback to BaseSourceOptions for AI-powered content relevance scoring
• NoopSynthesizer and stripMarkdownCodeFences remain in core

@debriefer/sources (1.0.1)

• Updated @debriefer/core dependency range for v2.0.0 compatibility

Package Scope Rename

All npm packages renamed to @debriefer org scope:

• debriefer -> @debriefer/core
• debriefer-sources -> @debriefer/sources
• NEW: @debriefer/ai

Other

• Package READMEs for all 5 npm packages + Python client
• Docker compose cleanup (removed unused Redis)
• CI improvements (cross-platform lockfile, cache fixes)

Migration

// Before
import { ClaudeSynthesizer, NoopSynthesizer } from "debriefer"

// After
import { NoopSynthesizer } from "@debriefer/core"
import { ClaudeSynthesizer, createAIDefaults } from "@debriefer/ai"

v1.0.1

Bump jsdom ^25→^28 and @mozilla/readability ^0.5→^0.6 to align with consumer projects and avoid duplicate transitive dependencies.

debriefer 1.0.0

Initial stable release of debriefer and debriefer-sources.

Packages

• debriefer (core) — Multi-source research orchestration engine with reliability scoring, phase-based execution, and AI synthesis
• debriefer-sources — Built-in research sources: Wikipedia, Wikidata, web search, news, archives, and more

Highlights

• Phase-based source execution with configurable early stopping
• Reliability scoring based on Wikipedia Reliable Sources Perennial list
• Cost tracking and per-actor cost limits
• Lifecycle hooks for observability
• Wikipedia person validation with AI date extraction
• Link following with configurable selectors
• 27 built-in sources across 8 phases

Production Proven

Running in production at deadonfilm.com since March 2026, processing death enrichment for 572,000+ actors.