Collection method: GitHub Trending bot issues from happydog-intj/ai-xiaohongshu-daily, comparing stars today in #560 (2026-10-14) against #559 (2026-10-13). Rain1er/trending (#405, #406) failed to fetch for the fourth run in a row, so this pick is based on happydog alone. Excluded: repos above 40k stars (morluto/rea at 201.6k, cathrynlavery/diagram-design at 53.9k), small gains (vercel-labs/skills, +187), and lyogavin/airllm (+1,004, up 14% day over day but only 2.6% of its total, below the new-entry bar).
mksglu/context-mode — +978 ⭐ today
https://github.com/mksglu/context-mode
Total stars: 28,062 · Previous day +246 → today +978 (+297.6% day over day; about 2.7x its last featured day, Oct 2, at +357) · TypeScript · Elastic License 2.0
Context Mode is an MCP server plus a set of hooks that saves context-window space for AI coding agents and keeps work state intact after compaction. Tool calls run in an isolated subprocess sandbox, and only stdout enters the context. Instead of pasting a 56KB Playwright snapshot or a 45KB log in full, it stores output in a per-project SQLite FTS5 index and retrieves only what is needed (the README claims 315KB down to 5.4KB, a 98% cut). Six hooks, including PreToolUse, PostToolUse, PreCompact and SessionStart, record file edits, git operations, tasks, errors and user decisions as events and restore them after compaction or resume. It supports 17 platforms, including Claude Code, Gemini CLI, Cursor, Codex CLI, Kiro, Zed and the Antigravity CLI, though hook support varies by platform. The jump from a quiet early October to a 4x daily gain likely reflects growing pain around context cost and forgetting in long-running agent sessions, plus its expanding platform list; we could not confirm a specific trigger.
Key features:
- Six sandbox tools (ctx_execute, ctx_batch_execute, ctx_index, ctx_search, ctx_fetch_and_index and more) plus five meta tools for stats and diagnostics
- Isolated subprocess execution across 12 language runtimes
- Hybrid search with BM25, trigram tokenizing, Reciprocal Rank Fusion, proximity reranking and typo correction
- Session continuity: tracks edits, git activity, tasks and decisions, then restores them after compaction
- Intent-based filtering for output over 5KB and automatic throttling of repeated searches
- URL cache with a 24-hour default TTL and automatic cleanup of data older than 14 days
Practical use: Drop it into long-running agent workflows that handle large logs or browser snapshots to cut token spend and context loss after compaction (note that ELv2 limits reselling it as a managed service).
Tags: #ContextEngineering #MCP #CodingAgents #ClaudeCode #SessionMemory #TokenOptimization
Summary
- No brand-new breakout repos today; context-mode returns after nearly quadrupling its daily gain.
- The top of the chart is still held by rea and diagram-design, both past 40k stars, while airllm keeps climbing steadily but stays below our new-entry bar.
Leave a Reply