v1.1.0

What's Changed

• feat: add themed Codex rendering, research rename, and search filters by @yuzu-octopus in #5
• Mateusdcc/gpt search output disrupts the tui in fullscreen by @mateusdcc in #7

New Contributors

• @yuzu-octopus made their first contribution in #5

Full Changelog: v1.0.0...v1.1.0

v1.0.0 - Native Web Research Harness

Release v1.0.0: Native Web Research Harness

v1.0.0 evolves simple web search into a full, model-independent Web Research Harness powered by OpenAI Codex's standalone search and browsing backend - with Zero Additional GPT Agent Turns and Zero GPT Tokens Spent.

Detailed Changelog

🚀 Web Research Harness Engine (web Tool)

• Multi-Action Web DTO (WebRunCommand): Added support for multi-query search (search_query), document content inspection (open), in-page pattern search (find), element/link interaction (click), and output length control (response_length: short | medium | long).
• Domain & Recency Filters: Models can constrain searches using domain whitelists (domains: ["github.com"]) and specify query recency windows.
• Document Reading (open): Enables models to inspect full page contents of search results using stable reference IDs (ref_id, e.g. turn0search0).
• Pattern Matching (find): Enables models to search for specific text patterns or code signatures directly inside opened document views.
• Persistent Session Continuity: Maintains stable session identity across multi-step research turns, preserving reference targets across sequential search -> open -> find -> click actions.

🎨 TUI, Progress & Citation Hyperlinks

• Live TUI Progress Updates: Integrated onUpdate handlers to report real-time status during research execution (e.g. Searching web..., Opening document turn0search0...).
• Collapsible Single-Line Status: Completed tool executions collapse into a single status line (✓ Web action complete (N results) (Ctrl+O to expand)), keeping the terminal clean.
• Clickable Terminal Hyperlinks (OSC 8): Converts private Unicode citation markers and turn IDs into clickable OSC 8 ANSI escape sequences (Cmd/Ctrl + click to open target URLs directly in default browser).
• Formatted Sources List: Appends a clean numbered Sources: index with hyperlinked titles and URLs at the bottom of web responses.

🛠️ Compatibility & Tools

• Single-Query Search Tool (web_search): Preserved backward compatibility for single-query lookups ({ query }), translating inputs directly into web({ search_query: [{ q: query }] }).
• Direct Terminal Slash Command (/gpt-search): Allows users to trigger direct web searches from the Pi prompt without spending LLM reasoning turns, outputting formatted result snippets (Title, URL, Snippet) with pagination support.
• Context Filtering (filterSearchContext): Automatically prunes system prompts, environment dumps, tool logs, and sensitive credentials before querying backend endpoints.

🧪 Comprehensive Test Suite (45 Tests Passing)

• Unit Tests (unit.test.ts, commands.test.ts, normalize.test.ts, output.test.ts, web-tool.test.ts): Validates DTO parsing, command validation, output formatting, context filtering, and collapsible TUI widgets.
• Integration Tests (provider-integration.test.ts): Mock transport tests for 200 OK, 401/403 auth errors, 429 rate limits, 500 server errors, timeouts, and manual cancellation.
• Live Endpoint Tests (real-endpoint.test.ts & real-search.test.ts): Exercises the live Codex endpoint for session continuity (search -> open -> find), response length controls, and domain filtering.
• Zero-GPT Network Interception (zero-gpt.test.ts): Verifies 0 GPT model inference calls (chat/completions, responses, turn/start) occur during web research commands.
• E2E Research Harness Suite (e2e-research.test.ts): End-to-end research tests running Pi CLI with active models (Gemini / Claude / DeepSeek), validating single-step search, multi-step search -> open, pattern matching, fresh lookups, and error recovery.

📚 Documentation Enhancements

• Restructured README.md: Follows familiar structure, clearly detailing /gpt-search terminal output examples, web_search LLM processing flow, and multi-step web harness execution steps.
• Technical Architecture (HOW-IT-WORKS.md): Updated system flow diagrams for both web_search (single-query) and web (multi-step research harness).

v0.1.0 - Initial Web Search Release

Initial release of pi-gpt-search providing standalone web search capability for Pi coding agent powered by OpenAI Codex standalone search engine.