All notable changes to openlens will be documented in this file.
openlens setupnow installs theusing-openlensskill for Claude Code and Codex- The skill teaches AI agents how to use the CLI: when to use each command, how to read output, what to do with results
skills/directory included in npm package
- Rewrote SKILL.md from command reference to workflow guide with patterns and decision logic
- README overview rewritten for clarity
openlens setupinteractive wizard with @clack/prompts for polished terminal UI- Setup wizard tests (
test/e2e/setup.test.ts) skills/using-openlens/SKILL.mdCLI usage reference for AI agents
- Updated all packages to latest: zod 4, hono 4.12, typescript 6, yargs 18, opencode SDK 1.3.3
- Fixed "OpenLens" casing to "openlens" lowercase throughout all docs, source, hooks, and plugins
- Node engine requirement updated to >=20.19
- Published to npm — install with
npm install -g openlens - Syntax highlighting in docs server (highlight.js)
- Client-side search across all wiki pages
- E2E tests for hooks install/remove (8 tests) and docs command (1 test)
repository,homepage,bugs,authorfields in package.json
- Version now read from package.json at runtime (single source of truth)
- README installation section: npm install is now the primary method
- Platform hooks trigger only on git commit/push (removed file-write hooks)
- Codex hook config changed from TOML to JSON (
hooks.json)
- CI lockfile mismatch (
bun.lockout of sync with package.json) - CLI help text wrapping at narrow terminal widths
- Wiki sidebar sort order (10-glossary now after 9-testing)
- Architecture tree in README: correct hook filenames, added missing files
- Removed self-dependency added during npm publish
- Confidence scoring: agents assess confidence per finding (high/medium/low), configurable threshold via
review.minConfidence - Context strategies: per-agent auto-gathering of relevant files (security, bugs, performance, style)
- Structured agent reasoning: 5-step methodology with few-shot examples
- Cross-agent verification with confidence-aware grouping
- GitHub inline PR comments on specific lines via Review API
- Incremental PR updates: resolved issues marked, progress shown on re-push
- Platform plugins: Claude Code (
/openlens), Codex CLI ($openlens), Gemini CLI (/openlens), OpenCode (native) openlens docscommand — serves wiki locally with dark theme and mermaid diagramsopenlens hooks install/remove— git pre-commit and pre-push hooks- Platform hook configs for Claude Code, Gemini CLI, Codex CLI
- DeepWiki documentation (11 pages with mermaid diagrams)
- How-to guides: CLI, plugins, CI/CD, hooks
- Live progress streaming during reviews (SSE events)
formatGitHubReviewlibrary export for GitHub review payloadsgatherStrategyContextandfilterByConfidencelibrary exports- 27 new e2e tests for v2 features (244 total)
- SSE stream reader — was falling back to polling, causing 180s timeouts on every review
- SARIF output corruption in CI — stderr was being redirected into the JSON file
- GitHub Action flag building — branch mode was producing malformed git diff arguments
- Base branch not available in CI checkout — added explicit fetch step
- Agent prompts updated with explicit domain boundaries to reduce cross-agent duplication
- Severity format enforced: must be exactly "critical", "warning", or "info"
- CLI help text shortened to prevent line wrapping
openlenshelp shows extended descriptions with quick-start examples
- Initial release
- CLI with run, agent, init, serve, models, doctor commands
- 4 built-in agents: security, bugs, performance, style
- SARIF, JSON, text, markdown output formats
- GitHub Actions composite action
- HTTP server with Hono
- Library API with 30+ exports
- OpenCode plugin integration
- Event bus for lifecycle events
- Suppression rules and .openlensignore
- Rules discovery (AGENTS.md, CLAUDE.md)
- CI auto-detection (GitHub Actions, GitLab, CircleCI, Buildkite, Jenkins, Travis)