Claude IS the AI — when running inside Claude Code, use Claude's own
language understanding for Stage 2 corrections instead of calling an
external API. No API key needed by default.
New capabilities in native mode:
- Intelligent paragraph breaks at logical topic transitions
- Filler word reduction (excessive repetition removal)
- Interactive review with confidence-level tables
- Context-aware judgment using full document context
API mode (GLM) remains available for batch/automation use cases.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Replace real Zhipu GLM API key with fake placeholder in mask_secret()
and SecretStr docstring examples. The real key was exposed in this
PUBLIC repo.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- add scrapling-skill with validated CLI workflow, diagnostics, packaging, and docs integration
- fix skill-creator package_skill.py so direct script invocation works from repo root
- fix continue-claude-work extract_resume_context.py typing compatibility for local python3
- bump marketplace to 1.39.0 and updated skill versions
Move verbose sections to references/ files, keeping concise pointers in
CLAUDE.md. Zero content loss — all documentation preserved in reference
files that Claude loads on demand.
Moved to references/:
- plugin-architecture.md (296 lines) — architecture docs
- plugin-troubleshooting.md (441 lines) — installation debugging
- new-skill-guide.md (241 lines) — detailed templates/checklists
- promotion-policy.md (60 lines) — third-party request policy
- youtube-downloader/references/internal-sop.md — yt-dlp SOP
Also fixed: Available Skills #36-42 indentation, deduplicated 4x
versioning sections into one, removed stale notes.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add skill to fix broken line wrapping in Claude Code exported .txt files.
Reconstructs tables, paragraphs, paths, and tool calls that were hard-wrapped
at fixed column widths.
Features:
- State-machine parser with next-line look-ahead
- Handles 10 content types (user prompts, Claude responses, tables, tool calls, etc.)
- Pangu spacing for CJK/ASCII mixed text
- 53 automated validation checks
- Safety: never modifies original files, verifies marker counts
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New scripts:
- fix_transcript_timestamps.py: Repair malformed timestamps (HH:MM:SS format)
- split_transcript_sections.py: Split transcript by keywords and rebase timestamps
- Automated tests for both scripts
Features:
- Timestamp validation and repair (handle missing colons, invalid ranges)
- Section splitting with custom names
- Rebase timestamps to 00:00:00 for each section
- Preserve speaker format and content integrity
- In-place editing with backup
Documentation updates:
- Add usage examples to SKILL.md
- Clarify dictionary iteration workflow (save stable patterns only)
- Update workflow guides with new script references
- Add script parameter documentation
Use cases:
- Fix ASR output with broken timestamps
- Split long meetings into focused sections
- Prepare sections for independent processing
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add 10-point high-quality PR formula based on real-world success cases
- Add investigation phase workflow (post to issue before PR)
- Add git history tracing techniques (git log, git blame)
- Add evidence-loop pattern (reproduce → trace → link → post)
- Add high-quality PR case study reference
- Update PR checklist with investigation steps
- Emphasize separation of concerns (detailed analysis in issue, fix summary in PR)
Key principles:
- Deep investigation before coding
- Minimal, surgical fixes
- Professional communication
- No internal/irrelevant details in PR
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Add _ensure_list_spacing() to handle lists without blank lines before them
- Modify _md_to_html() to preprocess markdown content via stdin
- Add automated test suite (scripts/tests/test_list_rendering.py)
- Fix: Lists without preceding blank lines now render correctly
- Original markdown files remain unmodified (preprocessing in memory only)
Root cause: Pandoc requires blank lines before lists per CommonMark spec.
Without preprocessing, lists following paragraphs render as plain text.
Tested scenarios:
✅ Lists with blank lines (normal case)
✅ Lists without blank lines (critical fix)
✅ Ordered lists without blank lines
✅ Original file integrity preserved
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
New skill that collects real financial data for any US publicly traded company
via yfinance. Outputs structured JSON with market data, historical financials,
WACC inputs, and analyst estimates. Includes 9-check validation script and
reference docs for yfinance pitfalls (NaN years, field aliases, FCF mismatch).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Update model ID to claude-sonnet-4-6 (latest, Feb 2026)
- Add Relay/Proxy API Configuration section with apiBaseUrl patterns
- Document that maxConcurrency MUST be under commandLineOptions (not top-level)
- Add LLM-as-judge relay provider config (apiBaseUrl not inherited)
- Add 5 new Troubleshooting entries from real-world title-agent eval
- Add Concurrency Control section to API reference
- Clarify file:// path resolution (always relative to promptfooconfig.yaml)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Move `npx @daymade/ccpm setup` to the top of SKILL.md so the agent
bootstraps the entire ecosystem before attempting any command.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Transform from passive user documentation to active Agent instructions
- Add allowed-tools: Bash, Read for direct command execution
- Add intent mapping table: user intent → ccpm command
- Add execution rules: Claude runs commands directly, no copy-paste
- Add MCP server cross-reference for Claude Desktop users
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add fourth conflict layer: SSH ProxyCommand double tunneling causing
intermittent git push/pull failures when Shadowrocket TUN is active.
Structural improvements per skill best practices:
- Eliminate content duplication between SKILL.md and reference
- Rename proxy_fixes.md → proxy_conflict_reference.md for clarity
- Trim SKILL.md from 534 to 487 lines (under 500 limit)
- Shorten YAML description from 910 to 661 characters
- Fix "apply all four" listing 5 items (separate anti-pattern)
- Clarify Layer 4's relationship to Tailscale theme
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add remote development SOP: proxy-safe Makefile patterns (--noproxy localhost),
SSH tunnel targets (tunnel/tunnel-bg with autossh), multi-port tunnels, and
end-to-end workflow with pre-flight checklist. Add diagnostic steps 2D (auth
redirect via SSH forwarding) and 2E (localhost proxy interception). Fix step
ordering, third-person description, and replace hardcoded IPs with placeholders.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add "move as-is, no compression" iron rule to prevent info loss during Level 2 migration
- Add anti-patterns 6 (compression during move) and 7 (disguising loss as "intentional deletion")
- Enhance Step 5 verification with 3 sub-checks: file existence, content completeness, no line counting
- Ban `wc -l` and line count mentions throughout the workflow
- Add real-world case studies 8 and 9 to principles reference
- Bump skill version to 1.2.0, marketplace to 1.32.1
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Bump claude-md-progressive-disclosurer 1.0.1 → 1.1.0 (missed in previous commit)
- Add mandatory "Updating Existing Skills" rule to CLAUDE.md: any commit
that modifies a skill's files MUST bump that skill's version in marketplace.json
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add windows-remote-desktop-connection-doctor v1.0.0 for diagnosing AVD/W365
connection quality issues with transport protocol analysis and log parsing
- Update claude-md-progressive-disclosurer SKILL.md and references
- Update marketplace to v1.32.0 (37 skills)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Key insight from real-world optimization:
- Optimizing once is not enough - need rules to prevent future bloat
- Add "信息记录原则" section to teach Claude what goes where
- This enables self-regulation: Claude knows Level 1 vs Level 2 placement
Changes:
- Add Principle 0: "信息记录原则" with template
- Update workflow Step 4: add "信息记录原则" as first step
- Update architecture diagram: show "信息记录原则" position
- Update quick checklist: add verification item
- Add Case 6 in references: missing self-regulation rules
Now "四条核心原则" instead of "三条核心原则":
0. 信息记录原则（防止未来膨胀）
1. 触发索引表放开头和末尾
2. 引用必须有触发条件
3. 代码模式保留在 Level 1
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add policy section defining what requests to decline
- Include rationale explaining scope creep, endorsement, and precedent concerns
- Add response template for polite but firm declines
- Document workflow for processing such requests
- Reference Issue #7 and PR #5 as precedent examples
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Bump skill-creator from v1.3.0 to v1.4.0
- Add complete YAML frontmatter reference table with all available fields
- Document `context: fork` for subagent-accessible skills (PR #6 by @costa-marcello)
- Add invocation control comparison table
- Document $ARGUMENTS placeholder with usage examples
- Add allowed-tools wildcard syntax examples
- Add hooks field inline example
- Update init_skill.py template with all optional fields
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Improvements to the YAML frontmatter documentation merged from PR #6:
- Add explanation of $ARGUMENTS placeholder and how it works with skill invocation
- Add inline example showing /deep-research invocation with arguments
- Expand allowed-tools field with more wildcard examples (npm, docker compose)
- Add hooks field example showing pre-invoke configuration
- Update init_skill.py template with wildcard examples and $ARGUMENTS hint
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add documentation for all available YAML frontmatter fields, with special
emphasis on `context: fork` which is critical for skills that subagents
should be able to use via the Task tool.
Changes:
- Add YAML Frontmatter Reference section to SKILL.md with complete field table
- Document when to use `context: fork` for subagent-accessible skills
- Add invocation control comparison table
- Update init_skill.py template with commented optional frontmatter fields
This addresses a gap where skills created without `context: fork` could not
be used by subagents, limiting the skill's usefulness in multi-agent workflows.
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Add speaker identification and pre-processing pipeline:
- Speaker identification via feature analysis (word count, segment count,
filler ratio, speaking style) with context.md team directory mapping
- New context_file_template.md for team directory configuration
- Intelligent file naming pattern: YYYY-MM-DD-<topic>-<type>.md
- Pre-processing pipeline with markdown-tools and transcript-fixer
- Transcript quality assessment workflow
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Add multi-tool orchestration for best-quality document conversion:
- Dual mode: Quick (fast) and Heavy (best quality, multi-tool merge)
- New convert.py - main orchestrator with tool selection matrix
- New merge_outputs.py - segment-level multi-tool output merger
- New validate_output.py - quality validation with HTML reports
- Enhanced extract_pdf_images.py - metadata (page, position, dimensions)
- PyMuPDF4LLM integration for LLM-optimized PDF conversion
- pandoc integration for DOCX/PPTX structure preservation
- Quality metrics: text/table/image retention with pass/warn/fail
- New references: heavy-mode-guide.md, tool-comparison.md
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Bump marketplace version: 1.24.0 → 1.25.0
- Bump skill-creator version: 1.2.2 → 1.3.0 (sanitization step)
- Add meeting-minutes-taker plugin entry
- Update skill count: 32 → 33 in all docs
- Update badges in README.md and README.zh-CN.md
- Add skill sections and use cases in both READMEs
- Update CLAUDE.md available skills list
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
Fixes#4
- Change npm package from @anthropic/ccpm to @daymade/ccpm
- @anthropic/ccpm does not exist (npm returns 404)
- @daymade/ccpm is the correct package (version 0.2.1)
- Add link to CCPM official website (https://ccpm.dev)
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
- Add skill-reviewer v1.0.0 for reviewing Claude Code skills against best practices
- Self-review mode: validate your own skills before publishing
- External review mode: evaluate others' skill repositories
- Auto-PR mode: fork, improve, submit PRs with additive-only changes
- Auto-install dependencies: automatically installs skill-creator if missing
- Add github-contributor v1.0.0 for strategic open-source contribution
- Four contribution types: Documentation, Code Quality, Bug Fixes, Features
- Project selection criteria and red flags
- PR excellence workflow and reputation building ladder
- GitHub CLI commands and conventional commit format
- Update marketplace to v1.22.0 with 30 skills
- Update documentation (README, README.zh-CN, CLAUDE.md, CHANGELOG)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Migrated all valuable architecture content from AGENTS.md into CLAUDE.md
- Created new 'Plugin and Skill Architecture' section in CLAUDE.md
- Includes: Core Concepts (Skills, Plugins, Agents, Commands)
- Includes: Architecture diagrams, installation flow, data flow
- Includes: Common misconceptions and best practices
- Deleted original AGENTS.md file
- Created symlink AGENTS.md -> CLAUDE.md to avoid duplicate maintenance
- Updated internal references to point to CLAUDE.md sections
This consolidation ensures single source of truth for plugin/skill architecture documentation.
- Add AGENTS.md: Complete architecture guide for Plugins, Skills, and Agents
- Explain the relationship between plugins and skills
- Document the GitHub-based marketplace mechanism
- Detail installation flow and data flow
- Clarify common misconceptions
- Provide best practices for authors, maintainers, and users
- Update CLAUDE.md: Add "Plugin and Skill Troubleshooting" section
- Systematic debugging process with real-world examples
- Common errors with root cause analysis and solutions
- "Plugin not found" error (most common: forgot to push)
- Stale marketplace cache issues
- JSON syntax errors
- Step-by-step debugging checklist
- Debugging commands reference table
- File locations reference
- Common pitfalls with wrong/correct examples
- Real-world case study: macos-cleaner installation issue
- Advanced cache inspection techniques
Key learnings from macos-cleaner deployment:
- Claude Code marketplace is GitHub-based, not local-file-based
- Local changes invisible until git push
- Cache requires explicit update after GitHub changes
- installed_plugins.json is source of truth for installations
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
- Add twitter-reader v1.0.0 for Twitter/X content fetching via Jina.ai API
- Implement secure API key management using environment variables
- Support individual and batch tweet fetching with bundled scripts
- Include Python (fetch_tweet.py) and Bash (fetch_tweets.sh) scripts
- Update marketplace to v1.20.0 (26 → 27 skills)
- Update all documentation (README.md, README.zh-CN.md, CLAUDE.md)
- Security: Remove hardcoded API keys, enforce HTTPS, add validation
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
