Conversation
- Remove optional start/end arguments for processing specific row ranges - Always process all patents from target_patents table - Update investigating-database skill invocation to request 'next patent ID' - Simplify skill interface and improve user experience Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add references/ directory with instructions.md, examples.md, troubleshooting.md - Move detailed screening process from SKILL.md to references/instructions.md - Add comprehensive judgment examples in references/examples.md - Add troubleshooting guide for common screening issues - Update SKILL.md to reference new documentation structure - Improve skill consistency across targeting and screening phases Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove Output Management section from references/instructions.md - Remove 'Context window full' issue from troubleshooting.md - Simplify documentation by removing context window optimization rules - Focus on core screening workflow without context management complexity Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove detailed implementation steps from SKILL.md and keep only high-level structure (Purpose, Prerequisites, Skill Orchestration, State Management, References). Detailed instructions are now in references/instructions.md. Changes: - Remove Step 1: Automated Screening (73-126 lines) - Remove Step 2: Generate Summary Report (128-152 lines) - Remove Output Management section (154-161 lines) - Remove Output section (163-166 lines) - Remove Quality Gates section (168-175 lines) - Remove Examples section (179-188 lines) - Remove Troubleshooting section (190-194 lines) - Add Skill Orchestration section - Add State Management section - Simplify References section This makes SKILL.md more concise (104 lines vs 195 lines) and follows the same pattern as targeting/SKILL.md. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove duplicate Skill Orchestration and State Management sections. Also remove reference to JSON files in Final State. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove detailed operation list as it should be in references/instructions.md. This makes SKILL.md consistent with targeting skill pattern. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace direct MCP tool calls with skill invocations: - fetch-patent → google-patent-cli:patent-fetch skill - Remove JSON file save/load references - Update instructions.md, examples.md, and troubleshooting.md This aligns with the skill-based architecture where all operations go through skills rather than direct MCP tool calls. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
These scripts are no longer needed since investigating-database skill handles patent data fetching. All operations should go through skills rather than direct script calls. Deleted: - plugin/skills/screening/scripts/shell/fetch-patent.sh - plugin/skills/screening/scripts/powershell/fetch-patent.ps1 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Refactor screening process to use parallel agent teams: - Replace sequential processing with parallel agent teams - Add team creation and patent assignment steps - Each teammate processes patents independently - Main agent orchestrates and aggregates results Benefits: - Faster processing for large patent volumes - Better resource utilization - Scalable architecture Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Update step 1 and step 5 to use consistent **Action** format: - Step 1: Get Unscreened Patents - Step 5: Aggregate Results This matches the targeting skill's documentation style for skill invocations. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Move Phase 3 (Generate Summary Report) responsibility from screening skill to investigating-database skill: - screening skill: Orchestrates screening process and requests progress - investigating-database skill: Generates screening.md report Changes: - Update instructions.md Phase 3 to request report from skill - Update SKILL.md to reflect new responsibility - Update examples.md to show new workflow - Update Output section to clarify report generation Benefits: - Clearer separation of concerns - investigating-database owns all database-related operations - screening skill focuses on orchestration Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove Phase 3 (Generate Summary Report) entirely as report generation is investigating-database skill's responsibility. Changes: - Remove Phase 3 from instructions.md - Remove "Generate Report" step from SKILL.md - Remove screening.md from Output sections - Remove Example 5 (Output Report Structure) from examples.md - Update example numbering screening skill now only: - Phase 1: Preparation - Phase 2: Automated Screening (with parallel agents) investigating-database skill handles: - All database operations - Report generation Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove database-level validation from screening skill's Quality Gates and Validation Checklist. These are investigating-database skill's responsibility: Removed from Quality Gates: - [ ] Each judgment is one of: relevant, irrelevant, expired - [ ] Screened count matches (Relevant + Irrelevant + Expired) - [ ] Summary statistics are accurate Removed from Validation Checklist: - [ ] No duplicate entries in screened_patents - [ ] All judgments are valid: relevant, irrelevant, or expired - [ ] screening.md follows template structure - [ ] Top 10 patents table correctly ordered - [ ] Progress statistics accurate Remaining checks (screening skill's responsibility): - [ ] All target_patents processed - [ ] NO Legal Assertions Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove Quality Gates and Validation Checklist sections entirely as these checks should be handled by individual skills: - investigating-database skill: Database-level validations - constitution-reminding skill: Legal compliance - legal-checking skill: Legal assertion checking screening skill focuses solely on orchestration. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Focus on orchestration and parallel processing responsibilities. troubleshooting.md changes: - Remove database-level issues (investigating-database's responsibility) - Add team/agent coordination issues - Add rate limiting and parallelism issues - Add agent failure handling - Add skill availability issues - Remove: "No patents found", "Resume support", "Template errors", "Database locked" examples.md changes: - Update Example 1 to show parallel processing with teams - Add Example 4 for handling rate limits in parallel context - Remove detailed report structure example (investigating-database's responsibility) - Focus on orchestration patterns and agent coordination Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Restore important nuance from git history (a72ba3a): - "Even if the domain differs, KEEP if technology could serve as infrastructure or a common platform for the product" This clarifies that infrastructure/platform technologies should be kept even when they're in different domains than the target. Changes: - Update instructions.md judgment criteria with detailed exception - Update examples.md Patent 4 reason to reflect the nuance Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add comprehensive test cases for screening skill's new architecture: New test cases: 1. functional-parallel-screening.toml - Tests parallel processing with teams - Verifies 2 agents can screen patents concurrently - Checks database creation and screening completion 2. functional-resume-screening.toml - Tests resume capability after interruption - Verifies no duplicate entries are created - Simulates partial screening scenario 3. integration-full-screening.toml - Tests complete end-to-end workflow - Verifies correct judgment distribution - Tests 4 patents with expected outcomes Changes: - Remove functional-get-next-patent.toml (obsolete architecture) - Remove explicit database setup (investigating-database creates it) - Add checks for skill loading and invocation - Add checks for judgment distribution Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Remove outdated "Step 2" reference from triggering conditions. Add file-based trigger similar to targeting skill: - patents.db exists with target_patents table populated Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add triggering.toml test case to verify screening skill is triggered when asked to screen patents or remove noise. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add comprehensive ETL script example at the beginning of import-csv.md to prevent AI from using direct .import command which causes CHECK constraint violations. Changes: - Add WARNING block with CRITICAL notice about always using ETL process - Add complete Quick Start section with ready-to-use SQL script - Include patent_id cleaning: remove hyphens and spaces - Include country code extraction from patent_id - Include date validation and normalization - Explicitly forbid direct .import to target_patents table This should fix CSV import failures due to: - CHECK constraint violated: instr(patent_id, '-') = 0 - Column count mismatches - Data format inconsistencies Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Change triple double quotes (""") to triple single quotes (''') in
all TOML test cases. TOML requires ''' for multi-line strings.
Also update test_prompt to be more explicit with numbered steps
for functional-parallel-screening test.
This fixes the issue where yq was reading test_prompt as null,
causing AI to receive empty messages.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…tabase ETL
Changes:
1. Add parse-toml.py script to parse TOML test case files
- Extract test_prompt, timeout, setup entries, and checks
- Handle TOML triple-quote syntax (''')
2. Update runner.sh to support both YAML and TOML files
- Detect file type by extension
- Use parse-toml.py for .toml files
- Keep yq for .yaml/.yml files
3. Move TOOLS_DIR definition to top of runner.sh
- Fix issue where TOOLS_DIR was used before definition
4. Fix import-csv.md with complete ETL script example
- Add WARNING block about always using ETL process
- Add Quick Start section with ready-to-use SQL script
- Include patent_id cleaning: remove hyphens and spaces
- Include country code extraction from patent_id
5. Add extract-toml-test-prompt.sh helper script
Test results:
- triggering: PASS ✅
- functional-parallel-screening: PASS ✅
- investigating-database ETL now working correctly
- Patent IDs properly cleaned (KR-102637029-B1 → KR102637029B1)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Fix TOML structure to ensure yq can parse test_prompt correctly. Move test_prompt definition before [[setup]] sections. Issue: yq v4 could not parse test_prompt when it came after [[setup]] sections in TOML files. Python tomllib also had the same issue. Solution: Reorder TOML file so test_prompt comes before [[setup]]. Files modified: - functional-parallel-screening.toml - functional-resume-screening.toml - integration-full-screening.toml - triggering.toml This fixes the issue where yq eval '.test_prompt' returned null. Note: yq v4 DOES support TOML natively. The issue was file structure, not parser capability. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add performance-analyzer skill for analyzing skill-bench test logs - Extract timeline with tool usage, parameters, and results - Identify bottlenecks (file operations, gaps, failures) - Visual format showing elapsed time, tool names, and status - Optimize investigating-database skill - Simplify initialize-database.md (remove unnecessary links) - Add explicit warning to prevent reading SQL file content - Improve performance from 51s to 19s for database initialization - Add functional-various-patent-id test case - Test various patent ID formats (US, KR, JP, CN, WO, CA, HK) - Validate ETL transformation logic for Google Patents CSV format - Ensure month zero padding for US patent IDs - Update screening instructions for clarity Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove mise installation from Dockerfile (build-time) - Add mise installation to post-create.sh (container start-time) - Run mise trust after installation to pre-approve config - This allows user-specific mise configuration and reduces image size Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Change TEMP TABLE to regular TABLE to fix connection scope issue - SQLite TEMP TABLEs disappear after each sqlite3 process - Regular TABLE persists across steps in ETL pipeline - Add DROP TABLE IF EXISTS to handle existing tables - Remove redundant Step 5 and Step 7 (verification steps) - Step 5: COUNT(*) check provides no meaningful info - Step 7: Redundant verification with no value for AI execution - Update SKILL.md with database initialization logic - Fix functional-various-patent-id test case (remove duplicate patent_10 check) - Add log-analyzing skill for performance analysis Performance improvement: - Before: 106.73s (with TEMP TABLE retry loop) - After: ~56s average (47% faster, stable across 5 test runs) - Eliminates 22-second retry loop caused by TEMP TABLE connection scope Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add script-type setup execution to runner.sh for database initialization and CSV import before AI execution, eliminating redundant import attempts. ## Changes ### runner.sh - Add script-type setup execution in setup phase - Execute setup scripts in WORK_DIR with tools/ in PATH - Maintain backward compatibility with existing file content setups ### tools/setup-db.sh - New generic database setup utility - Supports `init` command: Create database schema - Supports `execute <file.sql>` command: Execute SQL file - Handles claude-plugin directory resolution in workspace ### screening TOML files - Convert CSV data to direct INSERT statements - Remove redundant test_prompt - Add setup scripts for database initialization - functional-parallel-screening: 2 patents - functional-resume-screening: 3 patents + 1 pre-screened ## Problem Solved Before: AI would detect empty database and attempt re-import - 20+ minute wait times due to redundant import operations - Database initialization was part of AI execution flow After: Database initialized during setup phase - 37.83s execution time (stable across multiple runs) - AI focuses only on screening, not import - Clear separation: investigating-database handles DB, screening handles screening ## Test Results ✅ functional-parallel-screening: PASS (37.83s) - Setup executes correctly - AI does not attempt redundant imports - Screening completes successfully Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
… test Modified the SQL query to return count of duplicates instead of duplicate rows, ensuring proper comparison with expected value '0'. Previous query returned empty result when no duplicates existed, causing test failure despite correct behavior. Change: - FROM: SELECT patent_id, COUNT(*) FROM ... HAVING COUNT(*) > 1 - TO: SELECT COUNT(*) FROM (SELECT patent_id FROM ... HAVING COUNT(*) > 1) Test results: All screening tests now pass (3/3) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Restructured evaluating skill to match targeting skill architecture: - Moved templates/ to assets/ directory - Created references/ with examples.md, instructions.md, troubleshooting.md - Removed scripts/ directory (now uses investigating-database skill) - Updated SKILL.md to reference new structure This aligns evaluating with other skills and enables database-driven patent selection instead of shell script-based approach. Changes: - Renamed: templates/evaluation-template.md → assets/evaluation-template.md - Added: references/examples.md (6 detailed examples) - Added: references/instructions.md (complete workflow guide) - Added: references/troubleshooting.md (10 common issues) - Removed: scripts/shell/next-evaluation-patent.sh - Removed: scripts/powershell/next-evaluation-patent.ps1 - Updated: SKILL.md (simplified, references new files) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…kills - Split investigating-database into investigation-preparing and investigation-recording - investigation-preparing: database initialization, CSV import, data retrieval - investigation-recording: screening results, claims, and elements recording - Remove deprecated investigating-database skill and test cases - Add patent-evaluator agent with investigation-recording skill - Update evaluating skill to use new split skills - Fix setup-db.sh to reference investigation-preparing SQL files - Add check-log-contains.sh tool for batch insert verification - Add test cases: functional-record-claims, functional-record-elements - Update test cases to use check-db-query.sh and setup-db.sh init Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove constitution-reminding as a separate skill - Embed constitution rules directly into relevant skills: - targeting: Element-by-Element Analysis, Search Query Optimization - screening: Element-by-Element Analysis, Risk-Averse Screening, Breadth of Published Applications - evaluating: Element-by-Element Analysis - prior-art-researching: Element-by-Element Analysis, Comprehensive Literature Coverage, Evidence-Based Reporting, Prior Art Cutoff Date, Search Query Optimization - Remove constitution_loaded checks from test cases - Delete constitution-reminding test cases - Remove "Constitution" skill loading from all skill prerequisites Benefits: - More efficient: No need to load constitution-reminding skill separately - Context-focused: Each skill has only the rules it needs - Reduced overhead: Less skill loading overhead Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Remove constitution-reminding skill references from all skill files - Remove constitution-reminding references from examples, troubleshooting, instructions - Remove constitution_loaded checks from all test cases - Clean up concept-interviewing, targeting, evaluating, screening, progress-reporting, claim-analyzing This completes the constitution embedding refactor by removing all dependencies on the separate constitution-reminding skill. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
…iewing - Remove 'Constitution skill must be loaded' prerequisite - Remove 'Load Constitution (MANDATORY)' section Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Change `0-specifications/specification.md` to `specification.md` for simpler path - Update all investigating-database references to appropriate skills: - investigation-preparing: for data retrieval operations - investigation-recording: for data recording operations - Update test cases, documentation, and skill files accordingly This simplifies the file structure and makes the skill split clearer.
…gent ## Major Changes ### New Components - Add patent-evaluator agent for single-patent evaluation - Add investigation-fetching skill for data retrieval operations - Rename progress-reporting to investigation-reporting for consistency ### Skill Pattern Improvements - Rewrite SKILL.md files using Claude Code's natural language style - Remove pseudo-API syntax (Agent tool parameters) in favor of declarative instructions - Clarify "Each subagent handles exactly one patent" pattern ### Database Skill Refactoring - Update references from investigation-preparing to investigation-fetching - investigation-fetching: data retrieval (SELECT operations) - investigation-recording: data recording (INSERT operations) - investigation-preparing: database setup (initialization, CSV import) ### Cleanup - Remove all troubleshooting.md files (not useful) - Remove all examples.md files (not useful) - Remove functional-no-spec.toml test case (redundant) ### Testing Fixes - Fix check-db-query.sh to support numeric comparisons (>0, >=5, <10, etc.) - Add agent_used check to evaluating test cases - Update test cases for clarity Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
## Changes ### Internal Skills Protection - Add `user_invocable: false` to investigation-fetching, investigation-recording, investigation-preparing - These skills should only be invoked by other agents/skills, not directly by users - Prevents bypassing the proper skill orchestration flow ### Test Fix - Update evaluating test prompt to include "evaluate the patent" keyword - This ensures the evaluating skill is properly triggered ### MCP Server Configuration - Remove MCP server definitions from patent-kit plugin.json - google-patent-cli and arxiv-cli have their own MCP definitions - Avoids duplicate MCP server registrations ### Investigation-Report Template - Rename template file for consistency ## Result Evaluating test now passes with proper skill flow: - evaluating skill is triggered correctly - patent-evaluator agent is launched - claims and elements are stored successfully Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
## Changes - Add "evaluate" keyword to functional-parallel-analysis test prompt - This ensures the evaluating skill is properly triggered - Test now passes with 2 patents processed in parallel ## Result - 2 patent-evaluator subagents launched simultaneously (Δ0.000s) - 46 claims and 12 elements stored successfully - Execution time: 108s for parallel processing Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Create patent-screener agent in plugin/agents/ with YAML frontmatter - Remove Element-by-Element Analysis from screening (belongs to evaluating) - Remove unused assets/screening-template.md - Add agent_used check to screening test cases - Clean up old agents/patent-evaluator directory Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove references/instructions.md from screening and evaluating - Content already exists in SKILL.md, just remove references section - Simplify skill structure by reducing file count Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Initial State: unscreened/unevaluated patents exist - Final State: no unscreened/unevaluated patents remain Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Not used by any external skill - Remove from investigation-fetching operation mapping and references Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Analyze and decompose ALL claims (both independent and dependent) - Record elements for each claim individually - Update summary report to show total elements across all claims Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix prettier-broken escaped quotes in check command - Use double quotes instead of escaped single quotes Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add explicit CRITICAL note: "EVEN FOR A SINGLE PATENT" - Prevent AI from skipping subagent for single patent case - All evaluating tests now PASS (3/3) Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Not used by any external skill - Remove from investigation-fetching operation mapping, examples, and workflow Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove get-patent-id.md (not used in actual workflow) - Remove get-statistics.md (not used in actual workflow) - Remove corresponding test cases - Update SKILL.md to remove references Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Not used in actual workflow (investigation-fetching/get-next-patent.md is used instead) - Remove from investigation-preparing SKILL.md references Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- google-patent-cli:patent-search is an MCP server, not a skill - Update SKILL.md and references/instructions.md to reflect this - Update test check to verify MCP tool invocation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- The skill exists at skills/patent-search/SKILL.md - Targeting should use the skill, which internally uses MCP tools - Test should verify skill invocation, not MCP tool invocation Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add CRITICAL rule to always use Skill tool for google-patent-cli skills - Prevent AI from using MCP tools directly instead of skills Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Show how to use Skill tool to load google-patent-cli:patent-search - Add request format examples for assignee and query searches - Prevent AI from using MCP tools directly Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add functional-analyze-patent test case for claim-analyzing - Update claim-analyzer agent with new instructions - Update investigation-recording SKILL.md Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove question-responder from production agent skills list - Add AskUserQuestion tool for production use - Restructure workflow with clear sequential steps - Add explicit warning about not skipping Step 2 - Use Skill tool to invoke question-responder in test mode - Improve feature coverage check instructions Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove scripts/shell and scripts/powershell directories - Keep SKILL.md and templates Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prettier 3.8+ natively supports TOML files, so the plugin is no longer needed. Simplify .prettierrc to use built-in parser. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Prettier 3.8+ natively supports TOML files, so the plugin is no longer needed. Simplify .prettierrc to use built-in parser. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
sonesuke
force-pushed
the
feat/screening-sqlite-migration
branch
from
962a494 to
8aa4806
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Reorganize screening skill to match targeting skill structure and remove row number arguments for simpler interface.
Changes
🗂️ Restructure screening skill to match targeting pattern
references/instructions.md: Detailed screening process and workflowreferences/examples.md: Usage examples and judgment examplesreferences/troubleshooting.md: Common issues and solutions✂️ Simplify skill interface
<start> - <end>or<start>syntax🔄 Update investigating-database skill references
database→investigating-database(correct skill name)Benefits
Documentation
New Files
plugin/skills/screening/references/instructions.md(345 lines)plugin/skills/screening/references/examples.md(120 lines)plugin/skills/screening/references/troubleshooting.md(170 lines)Removed Features
<start> - <end>,<start>)Backward Compatibility
🤖 Generated with Claude Code