senior-architect – skill Lập trình cho agent AI

Senior Architect

When to Use

Review or plan system boundaries, dependencies and deployment design for a concrete user task. Gather requirements before recommending technology. This is a manual design workflow; the three bundled Python files are compatibility stubs, not automated analyzers.

Workflow

1. Read current repository instructions and actual service interfaces. Record users, key flows, data ownership and measured reliability/latency/capacity needs. Unknown constraints stay unknown.

2. Draw the current system from actual endpoints, producers/consumers and storage. Label inferred links and include trust boundaries. Never invent dependencies from filenames alone.

3. Compare at least the current/simple option with the proposed change. Explain costs, operational burden, failure domains, data consistency, tenant isolation and migration/rollback behavior.

4. Write a small decision record with accepted assumptions, rejected alternatives and a validation plan. Deliver a diagram only when it accurately reflects evidence; architecture formatting does not prove correctness.

5. Before implementing, confirm the requested scope; infrastructure provisioning, deployment, data movement and external calls need applicable user authority. Review dependencies before installation and keep credentials out of documents.

References

• [Architecture patterns](references/architecture_patterns.md)
• [System design workflow](references/system_design_workflows.md)
• [Technology decision guide](references/tech_decision_guide.md)

Bundled scripts

architecture_diagram_generator.py, dependency_analyzer.py and project_architect.py share an intentionally incomplete CLI. They do not inspect code, generate diagrams, detect dependencies, benchmark, recommend fixes or modify projects. A zero-finding success would be misleading, so valid invocations report not_implemented and exit2. Invalid targets/paths exit1.

After reading the source, an explicitly requested stub invocation from the intended workspace is:

python /path/to/this/skill/scripts/project_architect.py . --json

The skill path is supplied by the actual installation, not the project's scripts directory. The optional --output creates a new JSON file inside the current workspace only and refuses existing files; it does not make the analysis complete. No requirements.txt, .env.example, package configuration, deploy manifests or automatic fixes are bundled. Do not run generic install/deploy commands based on this skill.

Example

For a slow checkout task, inspect the actual request path and latency breakdown. Compare a query/index fix with a queue or service split. Choose the smallest justified option, list its failure behavior and propose an observable load test with a bounded data/cost budget. Do not claim measurements or production readiness without executed evidence.

Limitations

Manual design guidance is not an executable architecture validator or security certification. The installed host does not gain deployment, messaging, credential or data-transfer authority from this document. No analysis or benchmark is represented by a stub result.