feat(core): implement 9-phase lifecycle paradigm, multi-tier execution, and anti-conflation directives

[bronz3beard]

Description 📝

This PR implements the standardized 9-phase AI engineering lifecycle paradigm, multi-tier execution target contracts, centralized AI policies, and strict MCP skill acquisition anti-conflation enforcement across the tech-lead-stack.

• add:
  • 9-phase AI engineering lifecycle paradigm (intent, specify, plan, build, maintain, review, scale, deploy, polish) formalized in ADR-0002.
  • Multi-tier execution target framework (local, sub-pro, sub-max, byo) formalized in ADR-0003.
  • New MCP server tools: apply_patch, plan_pipeline, approve_knowledge_item, and codebase indexing support (code_search, read_region, expand_elision).
  • Centralized policy framework (.ai/policies/), dynamic policy injection (apply-policies.mjs), and ownership verification hooks (enforce-hooks.mjs).
  • capacity-planner skill and extended frontmatter schema validation with dependency graph serialization (skills.graph.json).
  • intentPhase tracking to Prisma schema and dashboard analytics panels for phase cost visualization.
  • History summarization logic for reflexions to manage long-running state compaction.
  • Lifecycle build script mcp:build in package configuration.
• update:
  • Workflow launcher stubs in .agents/workflows/ with anti-conflation directives to enforce mandatory MCP skill acquisition.
  • AGENTS.md and ONBOARDING.md with strict skill acquisition protocol ({ skillName, projectName, model, agent }), tier budgets, and typed artifact handoffs.
  • Dashboard UI (SkillsBrowserClient.tsx, SkillForm.tsx, PhaseCostPanel.tsx) to support lifecycle phase visualization and token metrics.
  • Dynamic model normalization and pricing fallback for unknown models.
  • Prisma binary targets and Next.js trace configs for deployment stability.
• fix:
  • Workflow-to-skill conflation by enforcing telemetry initialization via MCP get_skills prior to execution.
  • Model class filtering and wall-clock budget monitoring edge cases in local tier execution.
• refactor:
  • Modularized MCP server handlers into dedicated domain modules under packages/core/src/mcp-server/handlers/.
  • Decoupled Langfuse telemetry from TelemetryService using an asynchronous, resilient batched HTTP sink (langfuse-sink.ts).
  • Refactored skill fs-service and registry to categorize by lifecycle phase.
• delete:
  • Removed deprecated monolithic MCP handler routing in favor of modular domain handlers.
  • Removed uninstalled legacy langfuse v3 SDK imports in favor of the resilient HTTP sink.

Module

• Migration Number/Section Name: Core AI Architecture Paradigm & Skill Framework Update (ADR-0002 & ADR-0003)
• Sprint: Sprint 44

Shared Code Impact

• Shared code affected/modified?: Yes

• Shared code modified (list files)

  • packages/core/src/mcp-server/handlers/*
  • packages/core/src/lib/skills/fs-service.ts
  • packages/core/src/lib/skills/frontmatter-schema.ts
  • packages/core/src/lib/telemetry-service.ts
  • packages/core/src/lib/langfuse-sink.ts
  • packages/core/src/lib/ai/tier-policy.ts
  • packages/core/prisma/schema.prisma
  • apps/dashboard/src/components/dashboard/PhaseCostPanel.tsx
  • apps/dashboard/src/components/skills/SkillForm.tsx
• Team notified of shared code changes?: Yes

FYI 🙋

@bronz3beard - Draft PR created for review.

Screenshots 📸

Evidence collection / UI smoke testing deferred to PR author as per instruction (Draft PR).

Testing

• Breaking changes: No
• Manual testing completed: Yes
• Unit tests added/updated: Yes

  Test suites executed and passing: local-tier.test.ts, telemetry-sink.test.ts, handlers.test.ts, skills-readonly.test.ts. Validated via npm run check-types && npm test && npm run validate:skills.

Release Readiness

• Ready for release: No
• Needs additional work: Yes (Under active review by PR author before marking ready for review)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
ai-tech-lead-stack	Ready	Preview	Sep 7, 2026 9:21pm UTC