Compress session state into a ≤50-token CKPT checkpoint.
npx skills add https://github.com/mattbaconz/signal --skill signal-ckpt
Please help me install this Agent Skill. Skill: signal-ckpt Source: https://github.com/mattbaconz/signal/tree/main/gemini-signal/skills/signal-ckpt Command: npx skills add https://github.com/mattbaconz/signal --skill signal-ckpt
Compresses the current session state into a compact checkpoint to drop prior history and save tokens.
Invoke the checkpoint by saying /signal-ckpt or using the word 'checkpoint' to generate a compact session state you can resume from.
High-intent search queries and answers about installing and using this skill.
You can compress conversation history by triggering a checkpoint with /signal-ckpt or the word "checkpoint". This collapses prior context into a ≤50-token state atom, dropping old history while preserving the next actionable item to free up token limits.
A session checkpoint is a compact, deterministic state block that captures essential conversation context. You need to create one during long-running, context-heavy tasks to prevent hitting token limits and preserve critical state for resumption.
Yes, you can automatically save session state every five turns by enabling SIGNAL-3 mode. This auto-triggers the checkpoint mechanism to compress ongoing context-heavy conversations into a durable state block without manual intervention.
Yes, the checkpoint format preserves the next actionable item when dropping prior history. It uses a deterministic CKPT format and collapse algorithm to ensure the compressed state stays within the 50-token limit while retaining essential task continuity.
The best way to manage token limits in long-running conversations is to generate a concise checkpoint. Compressing the session state into a ≤50-token block drops bulky prior history while retaining the essential state needed to resume the task.
boazcstrikeCompress conversational and code context using anchored iterative summarization.
novatech2210-cmdLoad STATE.md, detect checkpoints, and route to the next action.
nazrulsoftwaredevSuggest manual context compaction at logical task boundaries in long-running sessions.
softwaresaltCapture session state into a checkpoint file and compact conversation history.
