Features

Project and phase management

Vibeless organises work into projects and phases. A project holds the goal statement, tech stack, constraints, and non-goals for a codebase. Phases break the project into ordered segments, each with its own tasks, acceptance criteria, and status. Your AI agent reads this structure through MCP tools like vibeless_get_project_context and vibeless_get_phase_spec, so it always knows what phase is active and what the current boundaries are.

Knowledge graph capture

When you open a project, Vibeless walks the repository and builds a code topology graph. Nodes represent files, modules, functions, and database tables; edges represent imports, foreign-key relationships, REST call sites, and IPC bindings. The graph is updated on every file-watcher event. Use vibeless_topology_summary to get node and edge counts, and vibeless_topology_query to filter by kind or name, or to pull a node's neighbours.

Reactive drift detection

Vibeless watches the project directory for file changes. On each change it runs the drift engine — a deterministic pipeline with five detectors covering tech-stack drift, constraint violations, scope creep, phase-boundary violations, and decision conflicts. No LLM is involved in the check itself. Call vibeless_run_drift_check from your agent to get a DriftReport sorted by severity, or trigger it from the app UI.

Constraint and decision log

Every architectural constraint and agent decision is persisted. Your agent calls vibeless_check_constraint before taking a structural action to get relevant constraints and overlapping prior decisions. It calls vibeless_log_decision to record what it did and why. vibeless_list_decisions returns the log newest-first so you can audit the trail at any time.

Phase boundary proposal

When a phase is close to done, vibeless_propose_phase_boundary analyses the current tasks, deliverables, and recent file changes, then produces a machine-readable boundary definition. This gives your agent — and you — a concrete, reviewable artefact before you mark the phase complete.

Structured handoff digest

vibeless_structured_handoff_digest produces a full project handoff covering what is done, what is open, watch-outs, and the recommended next action. It incorporates the decision log and any recent drift findings. Use it when switching agents, ending a session, or handing a project to another developer.

Context selection

When your agent calls vibeless_declare_intent, Vibeless extracts the scope of the work described and loads the governance context — rules, specs, standards — most relevant to that scope. vibeless_get_context returns the resolved context for a given scope string. This keeps each agent call focused on what matters instead of flooding it with the entire project.

Workspace and multi-project linking

If you are managing several related repositories, you can group them into a workspace. Projects within a workspace can be linked with coupling levels (awareness, dependency, read, governance). vibeless_get_workspace_context returns the workspace nodes and relationships; vibeless_check_cross_project_constraint evaluates a proposed action against workspace-wide governance rules.

Adapter-aware formatting

The MCP server detects which AI editor is calling it (Cursor, Claude Code, Codex, Gemini, or a generic MCP client) and formats its responses accordingly — adjusting token budgets and output structure per editor. You can override the detected adapter by setting adapterId in the server config.