📦 @goodandready/dsh-context-lens
Semantic AST Code Skeletonizer & Test Log Condenser for DeepSeek Harness
🇬🇧 English •
🇷🇺 Русский •
🇨🇳 中文说明
⭐ If you like this plugin, please star it on GitHub — it shows me that the plugin is useful to you and motivates me to keep developing it.
🐛 If you find a bug or would like to request a feature, open a GitHub issue in any language — I will review your proposal and implement useful suggestions in a future plugin version.
|
⚡ Overview & Semantic Compression Focus
Note on DSH Core vs Plugin Responsibility:
Context budgeting, token counting, tool result pruning (head/middle/tail), and spilling long results to disk (dsh-spill-policy, dsh-compaction-tool-result-pruner, dsh-token-meter) are natively managed by the DSH core (>= 0.1.5).
dsh-context-lens focuses strictly on what the core does not do — semantic compression:
- AST Code Skeletons: Signatures, interfaces, exported types, and doc comments instead of whole function/method bodies across 9+ languages.
- Path-Based Focus: Full code for active editing targets while surrounding workspace files remain skeletonized.
- Test & Build Log Compression: Failures, errors, stack traces, and test summary instead of raw multi-megabyte terminal streams.
graph LR
subgraph AgentInvocation [Agent Tool Invocations]
CodeTool["context_lens_code(filePath / code)"] --> Lens[dsh-context-lens Semantic Engine]
LogTool["context_lens_log(text, auto: true)"] --> Lens
end
subgraph Lens [Semantic Processing]
Lens --> Focus{Path Focus Check}
Focus -->|File in Focus| FullCode[Direct Read: Full Implementation]
Focus -->|Surrounding Files| AST[Direct Read / AST Skeletonizer]
Lens --> LogCompress[Log Condenser: Errors, Stack & Summary]
end
subgraph Output [Optimized Context]
FullCode --> Agent[🤖 AI Agent Context]
AST --> Agent
LogCompress --> Agent
end
style AgentInvocation fill:#1e1e2e,stroke:#89b4fa,stroke-width:2px,color:#cdd6f4
style Lens fill:#181825,stroke:#cba6f7,stroke-width:2px,color:#cdd6f4
style Output fill:#11111b,stroke:#a6e3a1,stroke-width:2px,color:#cdd6f4
✨ Features
1. 🧬 Multi-Language AST Structural Skeletonizer
- Supported Languages: TypeScript, JavaScript, Python, Go, C/C++, Rust, Java, and SQL DDL.
- Structural Preservation: Retains imports, classes, structs, interfaces, exported types, function signatures, and doc comments while discarding inner implementation bodies.
- Multiline Signature Support: Seamlessly accumulates complex multiline generic arguments, return types, and parameter lists up to block delimiters.
- Pure Regex Implementation: Zero heavy native dependencies or binary parser overhead; runs lightning-fast across any platform.
2. 📋 Intelligent Test & Build Log Condenser
- Supported Test Runners & Tools: Jest, Vitest, Pytest,
node --test, Go test, Cargo, Webpack, Vite, TSC, Maven, Gradle.
- Targeted Extraction: Identifies and preserves critical error messages, stack traces, assertion differences (
Expected ... Received ...), and failure context windows.
- Successful Run Summarization: Automatically collapses passing runs (0 failures) into concise summary lines.
- Auto-Mode Discrimination: Evaluates command context and text signatures to ensure normal non-test outputs are preserved untouched for core pruning.
- 3 Aggressiveness Modes:
raw: Removes noise lines while keeping general execution order.balanced: Preserves failure sections with surrounding context windows (default).aggressive: Extracts strictly error lines and stack frames.
3. 🎯 Path Focus Management
- Designate specific files or folders as active working targets for the current task.
- Files in focus bypass AST skeletonization, returning complete implementation code.
- Focus state is strictly scoped per session (
sessionId).
🛠️ Agent Tools Reference (Consolidated to 2 Tools)
All tools strictly conform to the DeepSeek Harness tool specification by providing output.render returning structured ContentBlock[] arrays ([{ type: 'text', text: ... }]), ensuring 100% session stability.
| Tool Name | Action / Mode | Parameters | Description |
|---|
context_lens_code | action: "skeleton" (default) | filePath?, code?, language?, maxDepth?, sessionId? | Generates a clean structural AST skeleton directly from filePath (reading file without preliminary read_file token waste) or from code. If file is in active session focus, returns full implementation |
context_lens_code | action: "focus" | paths: string[], sessionId? | Sets active focus file paths or patterns for the current session |
context_lens_code | action: "get_focus" | sessionId? | Retrieves active focus paths for the session |
context_lens_code | action: "clear_focus" | sessionId? | Clears focus paths for the session |
context_lens_log | N/A | text (or log), mode?, maxLines?, auto?, command? | Compresses test and build output logs, preserving failures, stack traces, and summary. In auto: true, only runs on recognized test/build outputs |
⚙️ Configuration Schema
Configured via the DSH Settings page (plugins.row.config / plugins.item):
| Property | Type | Default | Description |
|---|
compressionMode | string | 'balanced' | Log compression aggressiveness (raw, balanced, aggressive) |
astSkeletonMaxDepth | number | 3 | Max depth for AST skeleton generation |
autoCompressThreshold | number | 4000 | Character threshold for auto-compressing test/build logs when invoking context_lens_log with auto: true (0 to disable) |
📄 License
MIT © GooDAnDReaDY
