- P1: ContributionList items use `title` field (matches json-ui renderer),
was `text` which rendered blank step names in reports
- P3: Add --example flag to generate-report.mjs (documented in README)
- Update all JSON examples in SKILL.md and report-format.md to use `title`
- P1: Add package.json with js-yaml dependency (ERR_MODULE_NOT_FOUND on clean install)
- P1: Resolve CLI paths against process.cwd(), internal paths against input file dir
(was using __dirname which caused double-nesting like tests/tests/...)
- P2: Fix test status: pass+skip mix now reports PASS instead of SKIP
(anyPassed replaces allPassed to handle conditional skip steps correctly)
Add actionbook-web-test skill — a declarative YAML-driven browser testing
framework for Claude Code. Includes:
- SKILL.md with full execution flow, selector strategy, and recovery logic
- YAML workflow format with 17+ assertion types
- json-ui report generation with per-test collapsible sections and
base64-embedded screenshots for every test (PASS/FAIL/SKIP)
- generate-report.mjs script for building report JSON from execution results
- Reference docs for workflow format, assertion types, and report structure
- Remove keywords mapping from frontmatter description, focus on skill capabilities
- Add explicit Step 1b (AI Reformat to Markdown) between fetch and metadata extraction
to fix P1: fetch returns plain text, not Markdown — extracting title/images from
plain text would fail without the AI conversion step
- Update Quick Reference table to reflect plain text → AI → Markdown flow
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
P1: Replace non-existent --format markdown with default readability
extraction. Fetch returns plain text; AI reformatting produces Markdown.
Updated SKILL.md, references/twitter-handling.md, references/troubleshooting.md.
P2: Replace non-existent --wait-idle flag with --wait-hint heavy
across TROUBLESHOOTING.md and all references.
Also: Rewrite description to focus on capability and use cases,
removing keyword-style trigger examples per reviewer feedback.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Keep description in golden structure: [what] + [when] + [examples],
all examples in English only.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Export web articles and social posts to Obsidian Markdown with optional
AI translation. Includes progressive disclosure references, Twitter/X
special handling, and cross-platform compatibility (Claude Code + Codex)
via agents/openai.yaml.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix Table and ResultsTable components to use renderI18n() instead of
String() for cell values, resolving [object Object] in rendered HTML
- Prepare json-ui package.json for npm publishing (publishConfig, repo,
license Apache-2.0, prepublishOnly hook)
- Add dependencies and setup fields to deep-research plugin.json
declaring @actionbookdev/cli and @actionbookdev/json-ui as required
- Update SKILL.md render step to prefer npx as primary method
- Add json-ui install step to README, update troubleshooting table
- Add Dependencies section to README with one-command install
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add Demo Gallery to README with two real demos (Musk tweets on X.com, WebAssembly 2026 ecosystem)
- Add Chinese Content Quality Guidelines to SKILL.md to ensure native Chinese writing instead of machine translation
- Restructure workflow: Actionbook API query is now Step 2 (always first), arXiv Advanced Search is Step 3
- Remove all Chinese text from README for English-only documentation
- Add Bilingual Tone Reference table with bad/good translation examples
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Restructure workflow: arXiv Advanced Search is now Step 2 (always first),
Google/Bing search is Step 3 (supplement). Even non-academic topics like
"WebAssembly ecosystem" benefit from arXiv papers for deeper technical
insights. This reinforces Actionbook's core differentiator — no other
tool can operate the arXiv search form with 40+ indexed selectors.
Steps renumbered: 1-Plan, 2-arXiv, 3-Google, 4-Actionbook API,
5-Deep Read, 6-Synthesize, 7-Generate, 8-Render, 9-Open, 10-Close.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace manual cp with `npx skills add actionbook/actionbook --skill deep-research -g`
as the recommended installation method. Verified the skill is discoverable
from both repo root and playground/deep-research/ paths.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Three-layer guidance ensures any AI reading this skill will use arXiv
Advanced Search for academic topics:
1. Topic Detection table links paper/academic keywords to Step 2b
2. Architecture section adds "Why Actionbook" comparison table
3. Step 1 adds search method decision table with MUST-use directive
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add Step 2b documenting how Actionbook's 40+ indexed selectors for
arXiv Advanced Search enable multi-field, multi-criteria academic
searches that WebFetch/WebSearch cannot do. Includes comparison table,
two concrete examples, and usage decision guide.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a Claude Code Skill that researches any topic using Actionbook
browser CLI and generates beautiful HTML reports via json-ui.
Key features:
- Uses `actionbook browser` for web browsing (not WebFetch/WebSearch)
- Queries `actionbook search`/`get` API for verified selectors before browsing
- Generates bilingual (en/zh) json-ui JSON reports
- Works both as a plugin and as a standalone skill
- Includes battle-tested error handling from real testing
- Documented ar5iv fallback chains and json-ui pitfalls
Files:
- skills/deep-research/SKILL.md: Complete self-contained workflow (9 steps)
- commands/analyze.md: /deep-research:analyze command definition
- agents/researcher.md: Research agent with Actionbook browser tools
- examples/sample-report.json: Full json-ui example report
- README.md: From-zero setup tutorial
- .claude-plugin/: Plugin manifest and marketplace entry
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add README with demo explanation and quick start guide
- Include comparison of Claude Code with/without Actionbook
- Show scraping results: 25 vs 194 companies scraped
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
sen