reactive-feature-dev

name	reactive-feature-dev
description	Complete feature development workflow for Reactive Agents. Orchestrates the correct sequence of skills for any feature type. Use at the start of any non-trivial implementation task to route to the right skills and follow the right order.
user-invocable	false

Reactive Agents Feature Development Workflow

Step 0: Route to the Right Skills First

Before writing any code, identify what kind of work this is and load the corresponding skill:

Is this kernel work (new phase, guard, meta-tool, custom kernel)?
  → Load: kernel-extension skill

Is this provider or adapter work (new provider, streaming, adapter hooks)?
  → Load: provider-streaming skill

Is this MCP work (docker-based tools, transport, container lifecycle)?
  → Load: mcp-integration skill

Is this a new service in an existing package?
  → Load: implement-service skill

Is this a new package?
  → Load: build-package skill (add-new-package)

For all cases, also load: agent-tdd (write tests first), effect-ts-patterns (required for all code).

Step 1: Read Before Writing

# 1. Confirm current architecture
cat AGENTS.md | head -100

# 2. Read the affected package's source entry point
# (see architecture-reference skill for the right first file per package)

# 3. Identify which dependency layers are affected
# Changes to upstream packages (core, llm-provider) require downstream rebuild

Do NOT skip this step. Writing code without reading the affected package first causes:

• Duplicate implementations of existing functionality
• Wrong dependency layer — editing a higher-layer package when the change belongs in a lower one
• Breaking existing patterns because you didn't see them

Step 2: Identify Dependency Impact

Check the build order before touching any package:

Layer 0: core → llm-provider (changes here require rebuilding EVERYTHING downstream)
Layer 1: memory, tools, guardrails, cost, identity, observability, interaction, prompts, eval, a2a
Layer 2: reasoning, verification, orchestration, gateway, reactive-intelligence
Layer 3: runtime → reactive-agents (facade)

If your change touches Layer 0 or 1, plan for cross-package type verification.

Step 3: Write the Failing Test First (agent-tdd)

Following agent-tdd skill:

// 1. Create test file with mandatory header
// Run: bun test packages/<pkg>/tests/<your-file>.test.ts --timeout 15000

// 2. Write ONE failing test that captures the desired behavior
// 3. Run it — confirm it fails with the right error
bun test packages/<pkg>/tests/<your-file>.test.ts --timeout 15000

// 4. Write minimal code to pass
// 5. Run again — confirm it passes
// 6. Add edge case tests, refactor

Step 4: Implement with Effect-TS Patterns

Following effect-ts-patterns skill:

• All data shapes: Schema.Struct — never interface or type { ... }
• All errors: Data.TaggedError — never throw new Error()
• All state: Ref.make / Ref.get / Ref.update — never let
• All async: Effect.tryPromise — never await
• All imports: .js extension on relative imports (Bun ESM)
• All service methods: readonly keyword

Step 5: Validate Before Committing

Run both validators:

# 1. Pattern compliance (anti-pattern grep)
# (from validate-build skill)
grep -rn "throw new" packages/<pkg>/src/ || echo "✅ No throw"
grep -rn "^interface " packages/<pkg>/src/ || echo "✅ No plain interfaces"
grep -rn "let " packages/<pkg>/src/ || echo "✅ No let"
grep -rn "await " packages/<pkg>/src/ || echo "✅ No raw await"

# 2. Full package test
bun test packages/<pkg> --timeout 15000

# 3. Typecheck
bun run typecheck --filter @reactive-agents/<pkg>

# 4. Build (if touching a built package)
bun run build --filter @reactive-agents/<pkg>

For kernel changes, also run review-patterns Category 9 (kernel-extension compliance).

Step 6: Update Documentation

Following update-docs skill:

# Check what changed
git diff --name-only HEAD

# Update AGENTS.md if: new package, new builder method, test count changed
# Update README.md if: new package, new provider, new capability
# Update docs site if: new feature needs a docs page
# Update .agents/MEMORY.md with new capabilities/patterns

Step 7: Commit and Add Changeset

# Stage specific files — never git add -A
git add packages/<pkg>/src/ packages/<pkg>/tests/

# Commit with present-tense description of what changed
git commit -m "feat(<pkg>): <what the change does>"

# Add changeset for version bump
bun run changeset
# Choose: patch (bug fix), minor (new feature), major (breaking change)

Dead Code Areas — Do Not Touch

These sections exist in the codebase but are disabled. Do not modify, re-enable, or "clean up":

Area	Location	Size
buildDynamicContext	packages/reasoning/src/context/context-engine.ts	~280 LOC
buildStaticContext	same file (LIVE — active static prompt builder)	~280 LOC
Dead text-assembly functions	packages/reasoning/src/kernel/capabilities/attend/context-utils.ts	~690 LOC total

These are preserved as reference. Any PR that modifies them will be rejected.

Quality Gates Before Any Commit

All of these must pass. No exceptions:

• No any types (including as any casts)
• No interface for data shapes (use Schema.Struct)
• No let for mutable state (use Ref)
• No await (use Effect.tryPromise or Effect.sync)
• No throw (use Data.TaggedError)
• All relative imports have .js extension
• All tests pass: bun test packages/<pkg> --timeout 15000
• Build passes: bun run build --filter @reactive-agents/<pkg>

Subagent Context Template

When dispatching a subagent for implementation work, include this context block:

Project: Reactive Agents (reactive-agents-ts)
Stack: TypeScript + Bun + Effect-TS (^3.10) + bun:test
Key rule: All data shapes use Schema.Struct, errors use Data.TaggedError, state uses Ref, async uses Effect.tryPromise
Tests: always --timeout 15000, always targeted file not global suite
Required skills to load: effect-ts-patterns, [task-specific skill from routing above]
Dead code: do not touch buildDynamicContext/buildStaticContext or context-engine.ts dead sections
Architecture: read AGENTS.md before making changes