dsh-sieve: context engineering & token optimization plugin for DeepSeek Harness (DSH) — tool output filtering, context pruning, progressive skill disclosure. 36% smaller payload in offline replay. DSH 上下文管理与 token 优化节省插件。

typescript cordis ai-agent llm llm-agent context-window deepseek context-management prompt-compression agent-tools llm-cost ai-coding token-optimization token-efficiency coding-agent context-engineering context-pruning deepseek-harness tool-output dsh-plugin

• Updated Nov 1, 2026
• TypeScript

A stone gate for the context window.⛩️

ai-agents llm token-usage context-compression context-engineering hermes-agent tool-output

• Updated Nov 13, 2026
• Python

Context-cost reduction for the pi coding agent: lossless tabular re-encoding, structural reduction, cross-turn dedupe, opt-in history pruning and per-task skill disclosure — reversible, gated on never-worse, with honest metering.

typescript dedupe pi ai-agent llm prompt-caching context-compression token-optimization context-engineering pi-coding-agent pi-extension tool-output test-logs

• Updated Oct 18, 2026
• TypeScript

Evidence-preserving tool output compression for DeepSeek Harness — deterministic, local-first, and benchmarked.

dsh local-first ai-agent deepseek token-optimization coding-agent context-engineering deepseek-harness tool-output dsh-plugin

• Updated Aug 21, 2026
• HTML

Proof-of-concept: deterministic tool-output quarantine + best-effort prompt-injection pattern scan, hash-chained audit log

typescript audit-log ai-agents hash-chain ai-security quarantine guardrails prompt-injection llm-safety tool-output

• Updated Aug 20, 2026
• TypeScript

Local CLI for reducing context and tool-output noise in Claude Code, Codex, Gemini CLI, and other AI coding agents - without skipping verification.

cli privacy typescript ai developer-tools ai-agents local-first coding-agents gemini-cli agent-skills claude-code token-optimization codex-cli context-engineering ai-coding-agents tool-output

• Updated Oct 21, 2026
• TypeScript

Truncate text to a UTF-8 byte budget and a line budget without splitting characters.

cli typescript truncate text-processing utf8 ai-agents zero-dependencies head-tail llm context-window tool-output byte-length line-limit log-truncation

• Updated Oct 28, 2026
• TypeScript

Fail-closed compact renderer for structured LLM tool outputs: emit JSON, or switch to TOON columnar encoding only where it is measurably smaller under o200k.

agent json compression mcp toon llm token-optimization tool-output

• Updated Oct 16, 2026
• JavaScript

Local Codex plugin with resilient hook runtime recovery, compact tool-output receipts, and verification tracking.

verification developer-tools codex codex-plugin tool-output

• Updated Oct 3, 2026
• Python