mirror of
https://github.com/NousResearch/hermes-agent.git
synced 2026-05-02 08:47:26 +08:00
08b97660c5
Two features inspired by Claude Code's recent releases (v2.1.89–v2.1.101): 1. /context command (alias: /ctx) Shows a live breakdown of context window usage by component: - System prompt (identity, memory, skills index, context files, guidance) - Tool schemas (count and token estimate) - Conversation messages (by role: user, assistant, tool results) - Compaction summaries - Auto-compress threshold and remaining tokens - Visual progress bar This gives users visibility into what is consuming their context window, matching Claude Code's /context feature. 2. /compress <focus> — guided compression The existing /compress command now accepts an optional focus topic: /compress database schema When provided, the summariser prioritises preserving information related to the focus topic (60-70% of summary budget) while being more aggressive about compressing everything else. Inspired by Claude Code's /compact <focus> feature. Implementation details: - /context: new _show_context_breakdown() method in cli.py - /compress focus: focus_topic flows through _manual_compress → _compress_context → ContextCompressor.compress → _generate_summary, where it's appended to the LLM summarisation prompt - 15 new tests covering both features - No changes to prompt caching, message flow, or system prompt assembly