usewhisper-autohook
alinxus 作
Coding Agents & IDEs概要
Auto-hook tools for OpenClaw: query Whisper Context before every generation, ingest after every turn.
説明
このスキルでできること
Auto-hook wrapper for OpenClaw that queries Whisper Context before each agent generation and ingests the completed turn after. Defaults to delta compression and caching to minimize token use. Works with any stable user_id/session_id pair, with Telegram as the primary target.
Delta compression with locally persisted context hashes prevents re-sending unchanged memory on every turn, so token use stays lower than with full-history approaches.
使いどころ
- Persisting user context across Telegram bot sessions
- Recalling past decisions without replaying full chat history
- Injecting compressed long-term memory into agent prompts
- Reducing token costs on high-volume conversational agents
- Running a memory proxy for frameworks that always send full conversation history
ワークフロー例
User asks: What did we decide about the project timeline last week?
- 1
Calls get_whisper_context with user_id, session_id, and the current query
- 2
Receives compressed context string from Whisper Context API
- 3
Prepends context to the prompt as 'Relevant long-term memory: ...'
- 4
Generates response using the injected memory
- 5
Calls ingest_whisper_turn to store the user and assistant messages
Agent responds using past context; the turn is stored for future recall
要件
- Whisper Context API key from usewhisper.dev (WHISPER_CONTEXT_API_KEY)
インストール
clawhub install alinxus/usewhisper-autohookCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。