# Changelog

All notable changes to this project will be documented in this file.

The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).

## [0.3.2] - 2026-05-04

### Added

- Pipeline banner image and comparison infographic (`docs/images/01.png`, `docs/images/02.png`)
- `pi.image` in package manifest for [pi.dev gallery](https://pi.dev/packages/@curio-data/pi-intelli-search) preview
- Descriptive alt text for both README images (accessibility)
- Scannable Features list in README after intro paragraph
- Markdownlint config (`.markdownlint.yaml`)

### Changed

- Replaced Mermaid flowchart in README with rendered comparison infographic (02.png)

## [0.3.1] - 2026-05-04

### Added

- CI workflow (`.github/workflows/ci.yml`) — validates build, tests, and `npm pack` on every push/PR to `main`
- Release workflow (`.github/workflows/release.yml`) — publishes to npm on GitHub Release with provenance signing
- Release policy documented in AGENTS.md (explicit user permission required)
- E2E publish test (`test/run-e2e-publish.sh`) — installs from npm and validates the published package
- CI status badge in README

### Fixed

- `ensureCustomModels()` now creates `~/.pi/agent/` directory before writing `models.json` (fixes CI on fresh environments)
- `package-lock.json` drift corrected (version `0.1.0` → `0.3.0`, `@sinclair/typebox` → `typebox`)
- Release workflow uses correct `NPM_REPO` secret (was `NPM_TOKEN`)

## [0.3.0] - 2026-05-03

### Changed

- Updated copyright year to 2026 across all source files, NOTICE, README, and AGENTS.md
- Documentation now reflects 5-stage pipeline (added Stage 5: cache suggest) throughout README, ARCHITECTURE.md, and AGENTS.md
- Pi minimum version updated to ≥0.69.0 across all docs and badges (was incorrectly stated as 0.67.68/0.68.0 in README)
- `typebox` dependency correctly named in COMPONENTS.md and NOTICE (was `@sinclair/typebox`, migrated in v0.2.0)
- Removed spurious `thinkingLevelMap` compatibility entry from AGENTS.md (not used in codebase)
- Test count corrected to 104 across README and AGENTS.md
- Added npm downloads badge to README
- Pipeline diagram and cost table in ARCHITECTURE.md now include Stage 5 (cache suggest)
- Model configurability now documented in ARCHITECTURE.md provider choices section

## [0.2.0] - 2026-05-03

### Changed

- Updated pi SDK dependencies to 0.72.1 (`@mariozechner/pi-ai`, `@mariozechner/pi-coding-agent`)
- Migrated from `@sinclair/typebox` 0.34.x to `typebox` 1.x (pi ≥ 0.69.0 TypeBox migration)
- Minimum compatible pi version raised to 0.69.0

### Fixed

- Fixed `ERR_INVALID_URL` crash when Defuddle encounters pages with relative metadata URLs (e.g. GitHub `<link rel="canonical" href="/owner/repo/releases">`). Relative `href` and `content` attributes in `<meta>`, `<link>`, and `<a>` tags are now resolved to absolute URLs against the page URL before Defuddle processes the DOM.
- Fixed E2E test output verification — grep check no longer fails when the model doesn't echo the tool name in its response; cache artifact checks are the authoritative pass/fail

### Added

- 4 new unit tests for `cleanBrokenMetadata` covering relative URL resolution, literal undefined/null removal, and Defuddle integration with GitHub HTML
- E2E test (`test/run-e2e.sh`) now documented in AGENTS.md as a required step after every change
- `test/` source listing in AGENTS.md updated to include `run-e2e.sh`

## [0.1.0] - 2026-04-26

### Added

- 4-stage research pipeline: search → fetch → extract → collate
- `intelli_search` tool — web search via Perplexity Sonar (OpenRouter)
- `intelli_extract` tool — per-page LLM extraction with focus prompts
- `intelli_collate` tool — deduplication and synthesis into cached report
- `intelli_research` tool — full pipeline orchestrator (single call)
- Dual fetch strategy: Defuddle (HTML→markdown) vs raw markdown endpoint, with quality scoring
- Automatic `llms-full.txt` download for known documentation sites
- Persistent `.search/` cache with index, extractions, sources, and collated reports
- Perplexity Sonar model registration into `~/.pi/agent/models.json`
- Rate-limit monitoring via `after_provider_response` events with footer status
- Custom working indicator (🔍 🌐 📄 ✨) during pipeline execution
- Configurable settings via `~/.pi/agent/settings.json` and `.pi/settings.json`
- Agent-facing skill guide (`skills/intelli-search/SKILL.md`)
- 70 unit tests across 7 test files
- CI/CD via GitHub Actions (publish to npm on release)

[0.3.2]: https://github.com/Curio-Data/pi-intelli-search/releases/tag/v0.3.2
[0.3.1] https://github.com/Curio-Data/pi-intelli-search/releases/tag/v0.3.1
[0.3.0]: https://github.com/Curio-Data/pi-intelli-search/releases/tag/v0.3.0
[0.2.0]: https://github.com/Curio-Data/pi-intelli-search/releases/tag/v0.2.0
[0.1.0]: https://github.com/Curio-Data/pi-intelli-search/releases/tag/v0.1.0