lidge-jun/agentic-engineering

Agentic Engineering

Use this skill for engineering workflows where AI agents perform most implementation work and humans enforce quality and risk controls.

Operating Principles

1. Define completion criteria before execution.
2. Decompose work into agent-sized units.
3. Route model tiers by task complexity.
4. Measure with evals and regression checks.

Eval-First Loop

1. Define capability eval and regression eval.
2. Run baseline and capture failure signatures.
3. Execute implementation.
4. Re-run evals and compare deltas.

Task Decomposition

Apply the 15-minute unit rule:

• each unit should be independently verifiable
• each unit should have a single dominant risk
• each unit should expose a clear done condition

Model Routing

| Tier | Model | Use For | |------|-------|---------| | Fast/Cheap | Haiku 4.5 (claude-haiku-4-5) | Classification, boilerplate, narrow edits | | Balanced | Sonnet 4.6 (claude-sonnet-4-6) | Implementation, refactors, most coding | | Deep | Opus 4.8 (claude-opus-4-8) | Architecture, root-cause analysis, multi-file invariants |

Escalate model tier only when lower tier fails with a clear reasoning gap.

Session Strategy

• Continue session for closely-coupled units.
• Start fresh session after major phase transitions.
• Compact after milestone completion, not during active debugging.

Review Focus for AI-Generated Code

Prioritize:

• invariants and edge cases
• error boundaries
• security and auth assumptions
• hidden coupling and rollout risk

Focus review cycles on behavior and logic; rely on automated tooling for style enforcement.

Cost Discipline

Track per task:

• model
• token estimate
• retries
• wall-clock time
• success/failure

Escalate model tier only when lower tier fails with a clear reasoning gap.

Agent Development Lifecycle (ADLC)

Structured approach for building agent-powered features:

1. Define — Write eval criteria and acceptance tests before implementation
2. Build — Implement agent workflow with tool definitions (MCP preferred)
3. Test — Run capability evals + regression evals, measure quality metrics
4. Deploy — Canary rollout with observability; use feature flags for gradual enablement
5. Monitor — Track success rate, cost, drift, and user feedback in production

Tool Integration

• Use MCP (Model Context Protocol) as the standard for agent tool access
• Define tools with precise schemas — vague descriptions produce vague tool calls
• Prefer server-side tool execution over client-side when security matters

Agent Security & Governance

• Agents are non-human identities: apply zero-trust principles (least-privilege, time-bounded credentials)
• Audit every tool call and decision in structured logs
• Define a governance policy: which actions require human approval?
• Review agent outputs for compliance before external delivery

Team Structure

• Optimal: 2–4 person pods with AI agent support
• Each pod member reviews agent output, not just code
• Measure team by business outcomes, not vanity metrics (lines of code, commits)