Meronq
Meronq team

Sprint 6 closed: Skills and Workflows

sprint-6skillsworkflowsmcpadr

Sprint 6 is closed.

The goal was to introduce reusable procedural engineering knowledge — versioned Project Skills that offload deterministic workflows from LLM prompts into inspectable project code. See ADR-0006.

What Sprint 6 delivered

Every item on the roadmap is checked off:

  • Project Skill format — JSON schema with inputs, steps, safety, and evidence rules
  • project.build — run build via detected package manager
  • project.test — run test suite
  • adr.create — write a new ADR from template (title, optional context)
  • Workflow execution — sequential steps with fail-fast and structured results

Local MCP is now v1.16.0 with skill_list and skill_run. Product MCP and CLI expose the same skill surface.

New package: @meronq/skills

Module Role
listSkills() / getSkill() Built-ins + .meronq/skills/*.skill.json overrides
runSkill() Execute steps through injected SkillExecutionContext
buildAdrDraft() ADR numbering, slug, and markdown template

Built-in safety cannot be weakened by project overrides. Custom skills in v0 support create_adr steps only; command steps remain built-in exclusive.

Surfaces

meronq skill list
meronq skill run project.build
meronq skill run adr.create --title "My Decision" --context "Why we need this."

MCP tools: skill_list, skill_run. Handshake returns available_skills[] so assistants discover workflows without guessing shell commands.

Execution context lives in @meronq/server-core (createSkillContext) to avoid circular dependencies between skills and runtime.

Permission hardening

Post-sprint review added MERONQ_ALLOW_FILE_WRITE=true for skills that write files (e.g. adr.create) on the product MCP server. Runtime skills still require MERONQ_ALLOW_RUNTIME=true. Local MCP and CLI run skills without those gates for dogfooding and local use.

Pipeline

Intent (skill name + inputs)
  → getSkill / safety checks
  → createSkillContext (package manager, paths, runCommand)
  → sequential step execution
  → SkillRunResult (steps, output, action id)

Evidence recording into CEM/memory is planned; the schema is ready in skill definitions.

Sprint 7 — Agents

Sprint 7 goal: specialized engineering agents — architecture, code review, documentation, debugging, and onboarding — built on skills, memory, and MCP.

Procedures are now deterministic. Agents will decide when and why to invoke them.

Follow along

Sprint 5 made Meronq portable. Sprint 6 makes engineering procedures reusable across AI clients.

← Back to blog