Remplacement d'agent AMCOS

VérifiéPrudence

Remplace un agent IA défaillant lorsque la récupération est impossible ou que l'échec est terminal. Il confirme l'échec, préserve les artefacts récupérables dans un répertoire de transmission, demande l'approbation du gestionnaire via EAMA, crée un nouvel agent avec la compétence ai-maestro-agents-management, notifie l'orchestrateur (EOA) pour générer les documents de transmission et mettre à jour le kanban, et envoie la documentation validée au nouvel agent. Cette compétence est déclenchée par des événements de remplacement d'agent ou de basculement dans des environnements exécutant AI Maestro.

Spar Skills Guide Bot
DevOpsAvancé
10002/06/2026
Claude Code
#agent-replacement#failover#amcos#multi-agent#recovery

Recommandé pour

Notre avis

Cette compétence automatise le remplacement d'un agent IA défaillant dans un système multi-agents en coordonnant avec un gestionnaire, en préservant les artefacts de travail, en créant un nouvel agent et en transférant les tâches.

Points forts

  • Assure la continuité après une défaillance terminale
  • Préserve les artefacts de travail pour éviter toute perte
  • Fournit une liste de validation pour garantir l'intégrité du transfert
  • Définit clairement les rôles entre le créateur d'agents et l'orchestrateur

Limites

  • Nécessite l'approbation d'un gestionnaire externe (EAMA), ce qui peut introduire un délai
  • Dépend de l'infrastructure AI Maestro pour l'enregistrement des agents
  • Le nouvel agent n'a aucun souvenir de l'ancien, nécessitant une documentation de transfert complète
Quand l'utiliser

Utilisez cette compétence lorsqu'un agent IA a échoué de manière irrécouvrable et qu'un remplacement manuel est indésirable.

Quand l'éviter

Ne l'utilisez pas pour des défaillances temporaires où une reprise est possible, ou dans des configurations à agent unique.

Analyse de sécurité

Prudence
Score qualité92/100

The skill involves network requests (curl) to a local AI Maestro API for agent replacement notifications. It does not perform destructive actions, exfiltrate data, or disable safety mechanisms. However, the use of bash commands and network communication warrants caution in misconfigured environments.

Points d'attention
  • Uses curl commands to communicate with an internal API. While intended for local system messaging, improper configuration could expose the API. The commands include user-supplied agent names, which if not sanitized could lead to injection risks in naive implementations.

Exemples

Request Replacement Approval
Request replacement approval for a failed agent by sending an urgent message to the manager. Use the AMCOS replacement protocol: curl -X POST "$AIMAESTRO_API/api/messages" -H "Content-Type: application/json" -d '{"to":"eama-assistant-manager","subject":"URGENT: Replace libs-svg-svgbbox","priority":"urgent","content":{"type":"replacement-request","message":"Terminal failure (3 crashes). Requesting replacement approval.","failed_agent":"libs-svg-svgbbox"}}'
Notify Orchestrator of Replacement
Notify the orchestrator about a newly created agent so it can generate handoff docs and update the kanban. Use: curl -X POST "$AIMAESTRO_API/api/messages" -H "Content-Type: application/json" -d '{"to":"eoa-orchestrator","subject":"Agent replaced: libs-svg-svgbbox","priority":"high","content":{"type":"replacement-notification","message":"Please generate handoff docs and update kanban.","old_agent":"libs-svg-svgbbox","new_agent":"replacement-agent-uuid"}}'

name: amcos-agent-replacement description: Use when replacing a failed agent with a new instance. Trigger with agent replacement or failover events. user-invocable: false license: Apache-2.0 compatibility: Requires AI Maestro installed. metadata: author: Emasoft version: 1.0.0 context: fork agent: amcos-main

AMCOS Agent Replacement

Overview

Replace a failed agent when recovery has failed or failure is terminal. Covers failure confirmation, artifact preservation, manager approval, agent creation, orchestrator notification, and work handoff.

Prerequisites

  • Failure classified as Terminal, or all recovery strategies exhausted
  • AI Maestro running locally with agent registry accessible
  • Handoff directory at $CLAUDE_PROJECT_DIR/thoughts/shared/handoffs/

Instructions

Phase 4: Agent Replacement Protocol

  1. Confirm failure -- verify all recovery strategies attempted or failure is terminal
  2. Preserve artifacts -- save recoverable work to thoughts/shared/handoffs/AGENT_NAME/
  3. Request approval -- send urgent replacement request to EAMA; wait for approval
  4. Create replacement -- use ai-maestro-agents-management skill
  5. Notify EOA -- so it can generate handoff docs and update kanban
  6. Send handoff docs to new agent (validate with checklist below)
  7. Verify acknowledgment -- confirm new agent received handoff and began work
  8. Close incident -- log resolution, notify EAMA

See references/agent-replacement-protocol.md and references/op-replace-agent.md.

<!-- TOC: agent-replacement-protocol.md -->
  • Phase 1: Failure confirmation and artifact preservation
  • Phase 2: Manager notification and approval
  • ...and 8 more sections
<!-- /TOC -->

Replacement Protocol Summary

Terminal failure -> AMCOS notifies EAMA -> EAMA approves
  -> Create new agent -> Notify EOA (handoff + kanban)
  -> Send handoff to new agent -> Agent acknowledges

Key Consideration: Memory Loss

CRITICAL: The replacement agent has NO MEMORY of the old agent. Therefore:

  • EOA must generate handoff documentation
  • EOA must reassign tasks in GitHub Project kanban
  • AMCOS must send handoff docs to new agent

ROLE BOUNDARY: AMCOS creates agents and sends context. EOA owns task assignment.

Handoff Validation Checklist

- [ ] Required fields present (from/to/type/UUID/task/failed_agent/failure_reason)
- [ ] UUID is unique
- [ ] Target agent exists and is alive
- [ ] All referenced files exist
- [ ] No [TBD] placeholders
- [ ] Acceptance criteria defined

Output

| Phase | Output | |-------|--------| | Artifact preservation | Files saved to handoff directory | | Manager approval | Approval message from EAMA | | Agent creation | New agent online and registered | | Work handoff | Docs sent and acknowledged |

Error Handling

| Issue | Resolution | |-------|------------| | Manager does not respond | Wait 15 min, reminder, escalate to user | | New agent fails to register | Verify AI Maestro health and hooks | | Handoff rejected | Queue and retry in 5 minutes | | No artifacts recoverable | Document loss, notify EOA to recreate |

Examples

# Request replacement approval
curl -X POST "$AIMAESTRO_API/api/messages" \
  -H "Content-Type: application/json" \
  -d '{"to":"eama-assistant-manager","subject":"URGENT: Replace libs-svg-svgbbox","priority":"urgent","content":{"type":"replacement-request","message":"Terminal failure (3 crashes). Requesting replacement approval.","failed_agent":"libs-svg-svgbbox"}}'

# Notify orchestrator of replacement
curl -X POST "$AIMAESTRO_API/api/messages" \
  -H "Content-Type: application/json" \
  -d '{"to":"eoa-orchestrator","subject":"Agent replaced: libs-svg-svgbbox","priority":"high","content":{"type":"replacement-notification","message":"Please generate handoff docs and update kanban.","old_agent":"libs-svg-svgbbox","new_agent":"libs-svg-svgbbox-v2"}}'

Resources

Skills similaires