background-agent-pings
Background Agent Pings is a pattern for managing background tasks in Claude Code that uses system reminders instead of polling to track agent progress. Use this when launching long-running background agents to avoid wasting tokens on repeated status checks while continuing other work in parallel, relying on automatic notifications when the agent makes progress or completes output files.
git clone --depth 1 https://github.com/parcadei/Continuous-Claude-v3 /tmp/background-agent-pings && cp -r /tmp/background-agent-pings/.claude/skills/background-agent-pings ~/.claude/skills/background-agent-pingsSKILL.md
# Background Agent Pings
Trust system reminders as agent progress notifications. Don't poll.
## Pattern
When you launch a background agent, **continue working on other tasks**. The system will notify you via reminders when:
- Agent makes progress: `Agent <id> progress: X new tools used, Y new tokens`
- Agent writes output file (check the path you specified)
## DO
```
1. Task(run_in_background=true, prompt="... Output to: .claude/cache/agents/<type>/output.md")
2. Continue with next task immediately
3. When system reminder shows agent activity, check if output file exists
4. Read output file only when agent signals completion
```
## DON'T
```
# BAD: Polling wastes tokens and time
Task(run_in_background=true)
Bash("sleep 5 && ls ...") # polling
Bash("tail /tmp/claude/.../tasks/<id>.output") # polling
TaskOutput(task_id="...") # floods context
```
## Why This Matters
- Polling burns tokens on repeated checks
- `TaskOutput` floods main context with full agent transcript
- System reminders are free - they're pushed to you automatically
- Continue productive work while waiting
## Source
- This session: Realized polling for agent output wasted time when system reminders already provide progress updatesSecurity vulnerability analysis and testing
Agent Context Isolation
Agent Orchestration Rules
Agentic Workflow Pattern
Build Python agents using Agentica SDK - spawn agents, implement agentic functions, multi-agent orchestration
Guide for integrating Agentica SDK with Claude Code CLI proxy
Reference guide for Agentica multi-agent infrastructure APIs
Write reliable prompts for Agentica/REPL agents that avoid LLM instruction ambiguity