🚀 v3.5.0 Released — Grand Unification, Agent-Agnostic, 18 Sources, 20 Targets

[yusufkaraaslan]

🎉 Skill Seekers v3.5.0 Released — Grand Unification

TL;DR: One command. One interface. Direct converters. Agent-agnostic. All content extraction on by default.

Install / upgrade:

pip install --upgrade skill-seekers

📦 PyPI: https://pypi.org/project/skill-seekers/3.5.0/
🏷️ Release: https://github.com/yusufkaraaslan/Skill_Seekers/releases/tag/v3.5.0
📘 Website: https://skillseekersweb.com/ (docs, changelog, and feature matrix already updated for v3.5.0)

---

🚀 What's been done

This is one of the biggest refactors in the project's history — 80+ files changed — while staying fully backward-compatible for normal CLI usage.

🧩 Grand Unification (#346)

The create command is now the single source of truth for every source type. It auto-detects the input (URL, owner/repo, ./path, file.pdf, …), invokes the right converter directly, and runs centralized enhancement. No more thin wrappers, no more manual dispatch — just:

skill-seekers create https://docs.react.dev/
skill-seekers create facebook/react
skill-seekers create ./my-project
skill-seekers create manual.pdf

All 18 source types now route through the same pipeline with consistent flags, enhancement behavior, and error handling.

🤖 Agent-agnostic enhancement (#336)

Enhancement is no longer hard-coded to Claude. All 5 enhancers now go through a unified AgentClient abstraction and auto-detect based on which API key you have set. Supported out of the box:

Agent	Mode	Selection
Claude	API or CLI	default (ANTHROPIC_API_KEY)
Kimi	API or CLI	--agent kimi
Codex	CLI	--agent codex
Copilot	CLI	--agent copilot
OpenCode	CLI	--agent opencode
Custom	CLI	--agent-cmd "my-agent run"

The claude-enhanced merge mode was renamed to ai-enhanced — backward-compat alias kept, existing configs keep working.

🏪 Marketplace pipeline (#336)

• MarketplacePublisher — publish skills straight into Claude Code plugin marketplace repos
• MarketplaceManager — register and manage marketplace repositories
• ConfigPublisher — push configs to registered config source repos
• New push_config MCP tool for automated config publishing

🕷️ Smart SPA discovery (#336, #321)

Three-layer discovery engine: sitemap.xml → llms.txt → SPA nav rendering. Plus the new --browser flag uses Playwright to render JavaScript SPA sites (React, Vue, Next.js, etc.) that used to return empty HTML shells. Auto-installs Chromium on first use:

skill-seekers create https://tanstack.com/ --browser

Optional dependency: pip install "skill-seekers[browser]"

🩺 New doctor command (#316)

skill-seekers doctor

8 diagnostic checks: Python version, package install, git, core/optional deps, API keys, MCP server, output directory. Catches config/env issues before they hit the scraper.

🛡️ Prompt injection scanner (#324)

Bundled prompt-injection-check workflow now runs as the first stage in default and security-focus workflows. Scans scraped content for injection patterns (role assumption, instruction overrides, delimiter injection, hidden instructions) and flags suspicious content without removing it.

🆕 Other additions

• Kotlin language support ([FEATURE] Add Kotlin language support for codebase analysis #287) — full C3.x pipeline: AST, dependency extraction, GoF pattern detection (object→Singleton, companion object→Factory, sealed class→Strategy), JUnit/Kotest/MockK/Spek test extraction
• Codex CLI plugin manifest (feat: add Codex CLI plugin manifest #350) — ready-to-use OpenAI Codex integration
• All content extraction features enabled by default — pattern detection, test examples, how-to guides, config extraction, and router generation no longer require explicit opt-in. Use --skip-* flags to disable
• Dynamic routing via _build_argv() — added 7 previously-missing CLI flags
• 6 new behavioral UML diagrams — 3 sequence (create pipeline, GitHub+C3.x, MCP invocation), 2 activity (source detection, enhancement), 1 component
• 134 new tests — total now 3,194+ passing

🔧 Fixes worth calling out

• glob('*.md') → rglob('*.md') in all adaptors (fix: replace glob('*.md') with rglob('*.md') in all adaptors #349) — fixes packaging when skills live in nested directories
• GitHub languages API ([BUG] 目前所有版本通过GitHub生产skill都存在报错 #322) — filter non-integer metadata keys to prevent TypeError
• C3.x analysis ([BUG] C3.x codebase analysis doesn't work #323) — pass enhance_level instead of removed enhance_with_ai/ai_mode kwargs
• Path traversal protection in get_workflow_tool and workflow name parameter (Security: Path Traversal in workflow name parameter (CWE-22) #325, fix: prevent path traversal in workflow name parameter (CWE-22) #326)

🔐 Security

• Removed command injection via cloned repo script execution
• git add -A → targeted staging in marketplace publisher
• Auth tokens cleared from cached .git/config after clone
• defusedxml for sitemap XML parsing (XXE protection)
• Path traversal validation for config names

---

⚠️ Breaking changes?

Effectively none for normal usage. This is the short answer — a bigger refactor than usual but we went out of our way to keep things working:

Change	Breaking?	What to do
claude-enhanced merge mode → ai-enhanced	No — alias kept	Optional: update configs to new name
Content extraction features enabled by default	Behavior change — more output by default	Add --skip-patterns / --skip-examples / --skip-guides / --skip-config / --skip-router if you want the old output
50-file GitHub analysis limit removed	No — relaxes constraint	Larger repos now get full analysis
100-file config extraction limit removed	No — relaxes constraint	—
Unified scraper max_pages default 100 → 500	No — more permissive	Pass --max-pages 100 to keep the old cap
Enhancement timeout default 45min (unlimited supported)	No — more permissive	—
118+ hardcoded claude references removed internally	No — internal only	—

If you only use the CLI, your existing commands and configs should keep working. If you script against internal APIs, note that _call_claude_api → _call_ai in guide_enhancer.

---

📚 README & website — fresh for v3.5.0

• ✅ Core repo README — version badge, test count (2540+ → 3194+), source type count (17 → 18), platform count ("16 platforms" → "20 platforms (12 LLM + 8 RAG/vector)"), feature matrix, and architecture table all updated
• ✅ skillseekersweb — changelog, feature matrix, landing page, i18n translations (EN + ZH), meta descriptions, architecture reference, and introduction pages all updated for v3.5.0
• ✅ CHANGELOG.md — complete 3.5.0 entry with Grand Unification, agent-agnostic refactor, marketplace, SPA discovery, Codex plugin, rglob fix, and all security/CI items
• ✅ zh-CN docs — Chinese documentation updated in parallel

Full feature matrix: https://skillseekersweb.com/en/reference/feature-matrix/

---

📊 By the numbers

• 18 source types (web, github, pdf, word, epub, video, local, jupyter, html, openapi, asciidoc, pptx, rss, manpage, confluence, notion, chat, config)
• 20 output adaptors (12 LLM targets + 8 RAG/vector)
• 40 MCP tools
• 3,194+ tests passing
• 9 languages supported for code analysis (+ Kotlin in this release)
• 80+ files changed across this release

---

🙏 Thanks

Thanks to everyone who filed issues, tested pre-releases, and reviewed PRs for this one — especially the bug reports that caught #322, #323, and #349 before they hit a wider audience.

Try it out and let us know what you build:

pip install --upgrade skill-seekers
skill-seekers create your-favorite-docs-site.com

Questions? Drop them in Q&A. Found something broken? Open an issue. Want to show off what you built? Head to Show and tell 🚀