stasis-detection

name	stasis-detection
description	Use when deciding whether an agent or cron job should exit early due to inactivity. Governs the stasis detection pattern — cost-saving mechanism for idle agents/projects that suppresses cron runs when no progress has been made for 5+ consecutive days.
version	1.0.0
author	Hermes Agent
license	MIT
metadata	{"hermes":{"tags":["home-assistant","rocha-family","platform","cost-optimization","cron","governance"],"related_skills":["platform-manager","cron-patterns","hermes-governance","context-auditor"]}}

Stasis Detection

Overview

The stasis detection pattern prevents idle agents from consuming tokens unnecessarily. When a project, agent, or task domain has no new inputs, no progress, and no open work items for 5+ consecutive days, the agent logs the stasis event, increments a counter, and exits within ≤2 turns without doing work.

When to Use

• Any cron-driven agent that checks a project or domain
• Evaluating whether to skip a heartbeat run
• Platform manager nightly reflection — filter out stale agents
• Cost review — identify agents accumulating idle runs

Stasis Trigger Conditions

All of the following must be true:

1. stasis_consecutive_days >= 5
2. No new user input for the domain
3. No open tasks with approaching deadlines
4. No external events (API changes, alerts, calendar triggers)

If ANY condition is false → run normally.

Stasis Protocol (When Triggered)

1. Log: "Stasis day N — no new input. Skipping."
2. Append to events.log: {timestamp, agent, event_type: "stasis", stasis_day: N}
3. Increment stasis_consecutive_days in working.md
4. EXIT (≤2 turns total)
5. Do NOT send Telegram notification (silent skip)

Working Memory Format

Add this section to any domain's working.md:

## Stasis Tracking
- stasis_consecutive_days: 0
- last_active: YYYY-MM-DD
- stasis_triggered: false
- last_reset_reason: [brief description]

Reset Conditions

Reset stasis_consecutive_days = 0 when:

• New user message received for this domain
• External event triggers (bill due, appointment, etc.)
• Manual override from Hector
• Scheduled forced-run day (e.g., monthly reviews always run)

Agent File Format

Add this section to any agent's definition file:

## Stasis Detection
- Trigger: stasis_consecutive_days >= 5 AND no new input
- Action: log → increment → EXIT (≤2 turns)
- Reset: any new input or external event
- Never: send Telegram on stasis skip (silent)

Known Stasis Agents (Source Reference)

From htekdev/copilot-home-assistant standing-orders:

• carplay — 21+ days stasis
• milk-mama — 14+ days stasis

Hermes Application

Hermes should apply stasis detection to:

• teacher domain — formal curriculum suspended during Paula's recovery
• content-awareness — only active when OG flags content opportunity
• Any working.md with no updates in 7+ days → flag in context-auditor scan

Common Pitfalls

• ❌ Triggering stasis on family-critical domains (health, twins, Bella) — these always run
• ❌ Resetting counter without logging the reset reason
• ❌ Sending Telegram on stasis skip — it should be completely silent
• ❌ Applying stasis to heartbeat/mesh check-in — these always run regardless
• ❌ Counting weekends as stasis days for work-calendar agents

Verification Checklist

• Stasis counter in working.md for applicable agents
• Exit happens in ≤2 turns when triggered
• No Telegram sent on stasis skip
• Reset reason logged when stasis_consecutive_days resets to 0
• Critical-domain agents (health, NICU, dogs) excluded from stasis eligibility