adapt-architecture
À propos
Cette compétence propose une approche structurée pour faire évoluer progressivement l'architecture d'un système, en utilisant des modèles tels que la migration "strangler fig" et l'exécution en parallèle. Elle permet de réaliser des changements incrémentiels et sécurisés, comme des migrations de monolithe vers des microservices ou des remplacements de sous-systèmes, sans basculements en "big-bang". Utilisez-la lorsqu'une évaluation préalable a jugé le système prêt pour une transformation et que vous avez besoin d'une transition progressive avec la possibilité de revenir en arrière en toute sécurité.
Installation rapide
Claude Code
Recommandénpx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/adapt-architectureCopiez et collez cette commande dans Claude Code pour installer cette compétence
Documentation
Adapt Architecture
Execute structural metamorphosis — transform system architecture from current form to target form while keep operational continuity. Use strangler fig migration, chrysalis phases, interface preservation. System never stops working during transformation.
When Use
- Form assessment (see
assess-form) classified system as READY - System must evolve architecture to meet new requirements without downtime
- Migrating from monolith to microservices (or reverse)
- Replacing core subsystem while dependent systems keep operating
- Evolving data model while keep backward compatibility
- Any architectural change must be gradual, not big-bang
Inputs
- Required: Current form assessment (from
assess-formor equivalent analysis) - Required: Target architecture (what system should become)
- Required: Operational continuity requirements (what must not break during transformation)
- Optional: Available transformation budget (time, people, compute)
- Optional: Rollback requirements (how far back must retreat?)
- Optional: Parallel running duration (how long run old and new together)
Steps
Step 1: Design Transformation Blueprint
Plan metamorphosis path from current form to target form.
- Map transformation as sequence of intermediate forms:
- Current form → Intermediate form 1 → ... → Target form
- Each intermediate form must be operationally viable (serves traffic, passes tests)
- No intermediate form harder to maintain than current form
- Identify transformation seams:
- Where can current form be "cut" to insert new architecture?
- Natural seams: existing interfaces, module boundaries, data partitions
- Artificial seams: interfaces created specifically to enable cut (anti-corruption layers)
- Choose metamorphosis pattern:
- Strangler fig: new system grows around old, gradually replaces it
- Chrysalis: old system wrapped in new shell; internals replaced while shell preserves external interface
- Budding: new system grows alongside old; traffic gradually shifts (see
scale-colonyfor colony budding) - Metamorphic migration: phased replacement of components in dependency order (leaves first, roots last)
- Design interface preservation layer:
- External consumers must not experience disruption
- API versioning, backward-compatible contracts, adapter patterns
- Preservation layer is temporary scaffolding — plan its removal
Metamorphosis Patterns:
┌───────────────┬───────────────────────────────────────────────────┐
│ Strangler Fig │ New code intercepts routes one by one; │
│ │ old code handles everything else until replaced │
│ │ ┌──────────┐ │
│ │ │ Old ████ │ → │ Old ██ New ██ │ → │ New ████ │ │
│ │ └──────────┘ │
├───────────────┼───────────────────────────────────────────────────┤
│ Chrysalis │ Wrap old system in new interface; replace │
│ │ internals while external shell stays stable │
│ │ ┌──────────┐ ┌──[new]───┐ ┌──[new]───┐ │
│ │ │ old core │ → │ old core │ → │ new core │ │
│ │ └──────────┘ └──────────┘ └──────────┘ │
├───────────────┼───────────────────────────────────────────────────┤
│ Budding │ New system runs in parallel; traffic shifts │
│ │ ┌──────┐ ┌──────┐ ┌──────┐ ┌──────┐ │
│ │ │ Old │ │ New │ → │ Old │ │ New │ │
│ │ │ 100% │ │ 0% │ │ 0% │ │ 100% │ │
│ │ └──────┘ └──────┘ └──────┘ └──────┘ │
└───────────────┴───────────────────────────────────────────────────┘
Got: Transformation blueprint shows intermediate forms, seams, chosen metamorphosis pattern, interface preservation strategy. Each step concrete and testable.
If fail: No clean seam? System may need preliminary dissolution (see dissolve-form) to create seams before transformation. Intermediate forms not operationally viable? Transformation steps too large — decompose into smaller increments.
Step 2: Build Scaffolding
Construct temporary infrastructure that supports metamorphosis.
- Create anti-corruption layer:
- Thin translation layer between old and new systems
- Routes requests to appropriate system (old or new) based on migration state
- Translates data formats between old and new representations
- This layer is "cocoon" that protects transformation
- Set up parallel running infrastructure:
- Both old and new systems must be deployable together
- Feature flags control which system handles which traffic
- Comparison mechanisms validate that old and new produce equivalent results
- Establish rollback checkpoints:
- At each intermediate form, verify rollback to previous form possible
- Rollback must be faster than forward transformation step
- Data migration must be reversible (or data must be dual-written during transition)
- Build validation harness:
- Automated tests verify operational continuity at each intermediate form
- Performance benchmarks detect regression
- Data integrity checks catch migration errors
Got: Scaffolding infrastructure (anti-corruption layer, parallel running, rollback, validation) in place before any transformation begins. Scaffolding itself tested and verified.
If fail: Scaffolding too expensive? Simplify: minimum viable scaffolding is feature flag and rollback procedure. Anti-corruption layers and parallel running add safety but not always necessary for smaller transformations.
Step 3: Execute Progressive Cutover
Migrate functionality from old form to new form incrementally.
- Order components for migration:
- Start with least-coupled, lowest-risk component (build confidence)
- Progress toward more critical, more coupled components
- Save most coupled/critical component for last (by then team has experience)
- For each component: a. Implement new version behind anti-corruption layer b. Run parallel: both old and new process same inputs c. Compare outputs — should be equivalent (or differences should be expected and documented) d. When confident, switch traffic to new version (feature flag flip) e. Monitor for anomalies (increase monitoring sensitivity post-cutover) f. After stability period, decommission old version of this component
- Maintain continuous delivery throughout:
- Each cutover step is normal deployment, not special event
- System always in known, tested, operational state
- Cutover causes issues? Roll back to previous state (still operational)
Got: Functionality migrates component by component with validation at each step. System always operational. Each cutover builds confidence for next.
If fail: Parallel running reveals discrepancies? New implementation has bug — fix before cutting over. Cutover causes performance degradation? New component may need optimization or anti-corruption layer adding too much overhead. Team loses confidence mid-migration? Pause and stabilize — half-migrated system in known state far better than rushed full migration.
Step 4: Manage Chrysalis Phase
Navigate most vulnerable period — when system between forms.
- Acknowledge chrysalis reality:
- During migration, system is partly old and partly new
- This hybrid state inherently more complex than either pure state
- Complexity peaks at midpoint of migration, then decreases
- Chrysalis discipline:
- No new features during chrysalis phase (transformation only)
- Minimal external changes (freeze non-essential deployments)
- Increased monitoring and on-call coverage
- Daily check-ins on migration progress and system health
- Mid-chrysalis assessment:
- At halfway point, assess: target form still right goal?
- Anything changed (market, requirements, team) that affects target?
- Should transformation continue, pause, or redirect?
- Protect chrysalis:
- Keep rollback path clear at all times
- Document current hybrid state thoroughly (future debuggers will need it)
- Resist temptation to "clean up" temporary scaffolding before migration complete
Got: Chrysalis phase managed as deliberate, time-bounded period with increased discipline and monitoring. Team understands temporary complexity is cost of safe transformation.
If fail: Chrysalis phase drags too long? Hybrid state becomes new normal — worse than either old or new. Set time limit. Limit reached? Either accelerate remaining migration or accept hybrid state as "new form" and stabilize it.
Step 5: Complete Metamorphosis and Stabilize
Finish transformation. Remove scaffolding.
- Final cutover:
- Migrate last component(s) to new form
- Run full validation suite against complete new system
- Performance test under production-equivalent load
- Remove scaffolding:
- Decommission anti-corruption layer (no longer needed)
- Remove feature flags related to migration
- Clean up parallel running infrastructure
- Archive (don't delete) old system code for reference
- Post-metamorphosis stabilization:
- Run in new form for 2-4 weeks with enhanced monitoring
- Address any issues that emerge under real-world conditions
- Update documentation to reflect new architecture
- Retrospective:
- What went well in transformation?
- What was harder than expected?
- What would do differently next time?
- Update team's transformation playbook
Got: Transformation complete. System operates in new form. Scaffolding removed. Documentation updated. Team captured learnings for future transformations.
If fail: New form unstable after cutover? Maintain rollback path and continue stabilization. Stabilization takes more than planned period? Design issue in new architecture — consider whether targeted fixes or partial rollback of most problematic component appropriate.
Checks
- Transformation blueprint shows viable intermediate forms
- Scaffolding (anti-corruption layer, rollback, validation harness) in place before migration starts
- Components migrate in order from lowest to highest risk
- Parallel running validates equivalence at each step
- Chrysalis phase time-bounded with feature freeze discipline
- All scaffolding removed after transformation completes
- Post-metamorphosis stabilization period passes without critical issues
- Retrospective captures learnings
Pitfalls
- Big-bang migration: Transform everything at once. Abandons safety of incremental cutover. Maximizes blast radius. Always migrate incrementally
- Permanent scaffolding: Anti-corruption layers and feature flags never removed become technical debt. Plan scaffolding removal as part of transformation, not afterthought
- Chrysalis denial: Pretend hybrid state is normal. Leads to feature development on unstable foundations. Acknowledge chrysalis phase and enforce discipline
- Target fixation: Becoming so committed to target architecture that signs of better alternative ignored. Mid-chrysalis assessment exists for this reason
- Transformation fatigue: Long migrations exhaust teams. Keep each transformation step small enough to complete in days, not weeks. Celebrate milestones to maintain momentum
See Also
assess-form— prerequisite assessment determines if system ready for transformationdissolve-form— for systems too rigid to transform directly; dissolution creates seams needed hererepair-damage— recovery skill for when transformation introduces damageshift-camouflage— surface adaptation may suffice without deep architectural changecoordinate-swarm— swarm coordination informs sequencing of transformation across distributed systemsscale-colony— growth pressure is common trigger for architectural adaptationimplement-gitops-workflow— GitOps provides deployment infrastructure for progressive cutoverreview-software-architecture— complementary review skill for evaluating target architecture
Dépôt GitHub
Compétences associées
executing-plans
DesignUtilisez la compétence executing-plans lorsque vous disposez d'un plan de mise en œuvre complet à exécuter par lots contrôlés avec des points de contrôle de revue. Elle charge et examine le plan de manière critique, puis exécute les tâches par petits lots (3 tâches par défaut) tout en rapportant la progression entre chaque lot pour une revue par l'architecte. Cela garantit une mise en œuvre systématique avec des points de contrôle de qualité intégrés.
requesting-code-review
DesignCette compétence délègue un sous-agent réviseur de code pour analyser les modifications apportées au code par rapport aux exigences avant de poursuivre. Elle doit être utilisée après avoir terminé des tâches, implémenté des fonctionnalités majeures, ou avant une fusion vers la branche principale. La revue aide à détecter précocement les problèmes en comparant l'implémentation actuelle avec le plan initial.
connect-mcp-server
DesignCette compétence fournit un guide complet permettant aux développeurs de connecter des serveurs MCP à Claude Code via les transports HTTP, stdio ou SSE. Elle couvre l'installation, la configuration, l'authentification et la sécurité pour intégrer des services externes tels que GitHub, Notion et des API personnalisées. Utilisez-la lors de la configuration d'intégrations MCP, de la configuration d'outils externes ou du travail avec le Protocole de Contexte de Modèle de Claude.
web-cli-teleport
DesignCette compétence aide les développeurs à choisir entre les interfaces Web et CLI de Claude Code en fonction de l'analyse des tâches, puis permet une téléportation transparente des sessions entre ces environnements. Elle optimise le flux de travail en gérant l'état et le contexte de la session lors du passage entre le web, la CLI ou le mobile. Utilisez-la pour des projets complexes nécessitant différents outils à diverses étapes.
