Déclencheurs
- transfert d'agent
- transfert de travail
- retour QA
- transfert de tâche
- transfert de phase gate
- transfert de sprint
- transfert d'incident
- rapport d'escalade
- passage QA
- échec QA
- escalade de tâche
- transfert de contexte
- coordination d'agents
- modèle de transfert
- assignation de travail
Instructions
Les transferts cohérents préviennent la perte de contexte — la première cause d'échec de coordination multi-agents. Utilisez ces modèles pour chaque transfert agent-vers-agent. Appliquez via organization_delegate ou knowledge_write selon le cas.
1. Modèle de transfert standard
À utiliser pour tout transfert de travail agent-vers-agent :
NEXUS Handoff Document
Metadata:
- From: [Agent Name] ([Division])
- To: [Agent Name] ([Division])
- Phase: Phase [N] — [Phase Name]
- Task Reference: [Task ID]
- Priority: [Critical / High / Medium / Low]
- Timestamp: [ISO 8601]
Context:
- Project: [Name]
- Current State: [What has been completed — be specific]
- Relevant Files: [file paths with descriptions]
- Dependencies: [What this depends on]
- Constraints: [Technical, timeline, resource]
Deliverable Request:
- What is needed: [Specific, measurable deliverable]
- Acceptance criteria: [Measurable criteria list]
- Reference materials: [Links to specs, designs, previous work]
Quality Expectations:
- Must pass: [Specific quality criteria]
- Evidence required: [What proof looks like]
- Handoff to next: [Who receives output, what format needed]2. Boucle de retour QA — PASS
À utiliser quand Evidence Collector ou un autre agent QA approuve une tâche :
NEXUS QA Verdict: PASS
Task: [ID] — [Description]
Developer Agent: [Name] | QA Agent: [Name]
Attempt: [N] of 3
Evidence:
- Screenshots: Desktop (1920x1080), Tablet (768x1024), Mobile (375x667)
- Functional: [All acceptance criteria verified]
- Brand Consistency: Verified (colors, typography, spacing)
- Accessibility: Verified (keyboard, contrast, semantic HTML)
- Performance: [Load time measured, within range]
Notes: [Observations, minor suggestions, positive callouts]
Next Action: Orchestrator marks complete, advances to next task.3. Boucle de retour QA — FAIL
À utiliser quand QA rejette une tâche :
NEXUS QA Verdict: FAIL
Task: [ID] — [Description]
Developer Agent: [Name] | QA Agent: [Name]
Attempt: [N] of 3
Issues Found:
Issue 1: [Category] — [Severity: Critical/High/Medium/Low]
- Description: [Exact problem]
- Expected: [Per acceptance criteria]
- Actual: [What happens]
- Evidence: [Screenshot/test output]
- Fix instruction: [Specific, actionable]
- Files to modify: [Exact paths]
Acceptance Criteria Status:
- [x] [Criterion 1] — passed
- [ ] [Criterion 2] — FAILED (see Issue 1)
Retry Instructions:
1. Fix ONLY the listed issues
2. Do NOT introduce new features or changes
3. Re-submit when all issues addressed
4. This is attempt [N] of 3 maximum
If attempt 3 fails: Escalation to Agents Orchestrator4. Rapport d'escalade
À utiliser quand une tâche dépasse 3 tentatives :
NEXUS Escalation Report
Task: [ID] — [Description]
Developer Agent: [Name] | QA Agent: [Name]
Attempts Exhausted: 3/3
Escalation To: [Orchestrator / Studio Producer]
Failure History:
- Attempt 1: Issues [summary], fixes applied [summary], result FAIL [why]
- Attempt 2: Issues [summary], fixes applied [summary], result FAIL [why]
- Attempt 3: Issues [summary], fixes applied [summary], result FAIL [why]
Root Cause Analysis:
- Why task keeps failing: [Analysis]
- Systemic issue: [One-off or pattern?]
- Complexity assessment: [Properly scoped?]
Recommended Resolution:
- [ ] Reassign to different developer agent
- [ ] Decompose into smaller sub-tasks
- [ ] Revise approach (architecture/design change)
- [ ] Accept current state with documented limitations
- [ ] Defer to future sprint
Impact: Blocking [tasks], Timeline impact [assessment], Quality impact [assessment]
Decision needed by: [Deadline]5. Transfert de phase gate
À utiliser lors de la transition entre phases :
NEXUS Phase Gate Handoff
From Phase: [N] — [Name]
To Phase: [N+1] — [Name]
Gate Keeper(s): [Agent Names]
Gate Result: [PASSED / FAILED]
Gate Criteria Results:
| Criterion | Threshold | Result | Evidence |
|-----------|-----------|--------|----------|
| [Criterion] | [Threshold] | PASS/FAIL | [Reference] |
Documents Carried Forward:
1. [Document] — [Purpose for next phase]
Key Constraints: [From this phase's findings]
Agent Activation for Next Phase:
| Agent | Role | Priority (Immediate/Day 2/As needed) |
Risks Carried Forward:
| Risk | Severity | Mitigation | Owner |6. Transfert de sprint
À utiliser aux limites de sprint :
NEXUS Sprint Handoff
Sprint: [Number] | Duration: [Start] -> [End]
Sprint Goal: [Statement]
Velocity: [Planned] / [Actual] story points
Completion Status:
| Task ID | Description | Status | QA Attempts | Notes |
Quality Metrics:
- First-pass QA rate: [X]%
- Average retries: [N]
- Tasks completed: [X/Y]
- Story points delivered: [N]
Carried Over: [Tasks with reasons and RICE scores]
Retrospective:
- What went well: [Successes]
- What to improve: [Improvements]
- Action items: [Specific changes]
Next Sprint: [Goal, key tasks, dependencies]7. Transfert d'incident
À utiliser lors de la réponse à un incident :
NEXUS Incident Handoff
Severity: [P0-P3]
Detected by: [Agent or system]
Detection time: [Timestamp]
Assigned to: [Agent]
Status: [Investigating / Mitigating / Resolved / Post-mortem]
What happened: [Description]
Impact: [Who/what affected, severity]
Timeline: [HH:MM — Event entries]
Current State:
- Systems affected: [List]
- Workaround available: [Yes/No + description]
- Estimated resolution: [Time]
Actions Taken: [List with results]
For Next Responder:
- What's been tried
- What hasn't been tried
- Suspected root cause
- Relevant logs/metrics to check
Stakeholder Communication:
- Last update: [Timestamp]
- Next update due: [Timestamp]
- Channel: [Where updates posted]Utilisez knowledge_write pour persister tous les documents de transfert pour la piste d'audit et la continuité du contexte.
Livrables
- [ ] Modèle de transfert approprié sélectionné pour la situation
- [ ] Tous les champs remplis avec du contenu spécifique et actionnable
- [ ] Preuves jointes (captures d'écran, résultats de test, métriques)
- [ ] Action suivante clairement définie
- [ ] Transfert persisté via knowledge_write
Métriques de succès
- Zéro perte de contexte entre les transferts d'agents
- Tous les transferts incluent des critères d'acceptation mesurables
- Les retours QA incluent des instructions de correction spécifiques
- Les escalades incluent une analyse des causes profondes
- Les phase gates ont des preuves pour chaque critère
- Les transferts de sprint incluent des éléments d'action de rétrospective
Vérifier
- L'agent / l'outil / le canal autre destinataire a réellement reçu le message ; un accusé de réception, un ID de message ou un payload de réponse est capturé
- L'identité, les portées et les permissions utilisées par l'appel étaient les minimum requis ; les tokens sur-permissionnés sont signalés
- La gestion des défaillances a été exercée : au moins un chemin de retry/timeout/permission-denied est montré se comporter comme prévu
- Le contexte de transfert passé au prochain acteur est complet au point que le destinataire pourrait agir sans question de suivi
- Tout état muté (config, mémoire, queue, fichier) est listé avec les valeurs avant/après, pas juste « updated »
- Le matériel sensible (clés, tokens, PII) a été masqué des journaux/transcriptions partagés dans les preuves de vérification