Monitor AI interactions for context degradation using name and turn counter signals.
npx skills add https://github.com/apexradius/agentic-os --skill context-canary
Please help me install this Agent Skill. Skill: context-canary Source: https://github.com/apexradius/agentic-os/tree/main/skills/context-canary Command: npx skills add https://github.com/apexradius/agentic-os --skill context-canary
This Skill detects the degradation of context in AI interactions, signaling when the AI is no longer effectively attending to earlier instructions, which can lead to loss of context, instruction drift, or compaction loss.
Use the context-canary skill with the 'Sam' name to monitor context degradation during an AI interaction.
High-intent search queries and answers about installing and using this skill.
A context canary detects context loss by requiring the AI to include a specific name and turn counter in every response. If the AI stops including this signal, it indicates the ongoing context and earlier instructions have degraded.
Monitor AI context safety during long or high-stakes sessions where instruction drift or compaction loss poses a significant risk. This ensures the AI adheres to ongoing context and earlier instructions without degradation.
Recover context when an AI stops following instructions by triggering a trip protocol. This protocol handles missing canary signals through checkpointing and re-installing the canary to restore adherence.
Instruction drift occurs when an AI stops attending to earlier instructions during long sessions. A canary signal prevents it by embedding a required name and turn counter in responses, alerting you when context is lost.
Yes, you can use context monitoring for compaction loss in AI agents. The canary signal detects when the agent no longer effectively attends to earlier instructions and initiates a recovery protocol.
jason-c-devConfigures CloudWatch, CloudTrail, AWS Config, and Security Hub across accounts and regions.
TumeloRamaphosaCapture git state, decisions, and remaining work into checkpoint artifacts.
paolomozManage user session context across multi-turn interactions with sessionStorage and KV caching.
amanikMonitor live production apps for console errors, latency regressions, and page failures.
