How brifl/coding-agent-orchestration works
Vibe is a coding-agent orchestration system: a user asks the dispatcher for status or the next action, follows a selected prompt loop, and records verified progress in file-backed state until checkpoints and stages are complete. The graph emphasizes the dispatcher, prompt catalog, workflow sequencing, planning and checkpoint support, and installed skill experience. It also preserves the RLM context-bundle/runtime/provider branch. Source excerpts are partial; unsampled feature links are shown only where the README or SOURCE INDEX supports them.
Main entry points: agentctl.py Workflow CLI · prompt_catalog.py Prompt lookup · context_bundle.py Context bundler · PLAN.md Checkpoint plan
Components 25
Workflow dispatch
- Workflow CLI
tools/agentctl.py - Workflow sequencing
tools/workflow_engine.py - Configured loops
workflows
Planning and state
- Workflow state
.vibe/STATE.md - Checkpoint plan
.vibe/PLAN.md - History and context
.vibe - Plan generation
tools/plan_pipeline.py - Checkpoint templates
tools/checkpoint_templates.py
Prompts and skills
- Loop prompt catalog
prompts/template_prompts.md - Prompt lookup
tools/prompt_catalog.py - Prompt path resolution
tools/prompt_catalog_paths.py - Resource resolution
tools/resource_resolver.py - Agent skills
.codex/skills - Repository bootstrap
tools/bootstrap.py - Skill management
tools/skillctl.py
RLM task runtime
- Context bundler
tools/rlm/context_bundle.py - RLM task validation
tools/rlm/validate_task.py - RLM skill executor
skills/rlm-tools/executor.py - Persistent RLM runtime
tools/rlm/runtime.py - Provider interface
tools/rlm/providers/base.py - Provider checks
tools/rlm/provider_check.py - HTTP helpers
tools/rlm/providers/http_utils.py - Bundle and run files
Other components
- Engineer / agent
- LLM endpoint
Connections 27
- Engineer / agent → Workflow CLI: requests next action
README.md - Workflow CLI → Workflow state: reads and updates
README.md - Workflow CLI → Checkpoint plan: consults backlog
README.md - Workflow CLI → Workflow sequencing: uses sequencing
tools/agentctl.py - Workflow sequencing → Configured loops: loads loops
- Workflow CLI → Prompt lookup: retrieves prompt
tools/agentctl.py - Prompt lookup → Prompt path resolution: resolves catalog
tools/prompt_catalog_paths.py - Prompt path resolution → Resource resolution: finds installed resource
tools/prompt_catalog_paths.py - Prompt lookup → Loop prompt catalog: reads prompts
- Engineer / agent → Loop prompt catalog: follows selected loop
README.md - Engineer / agent → Workflow state: records evidence
README.md - Repository bootstrap → Agent skills: installs skills
README.md - Skill management → Resource resolution: resolves resources
tools/skillctl.py - Engineer / agent → Repository bootstrap: bootstraps repository
README.md - Workflow CLI → Plan generation: generates plan
tools/agentctl.py - Plan generation → Checkpoint plan: writes checkpoints
- Workflow CLI → Checkpoint templates: instantiates template
tools/agentctl.py - Checkpoint templates → Checkpoint plan: adds checkpoint
- Workflow CLI → History and context: captures context
README.md - Context bundler → RLM task validation: validates task
tools/rlm/context_bundle.py - Context bundler → Bundle and run files: writes bundle
- RLM skill executor → Context bundler: builds context
skills/rlm-tools/executor.py - RLM skill executor → Persistent RLM runtime: runs task
skills/rlm-tools/executor.py - Persistent RLM runtime → Bundle and run files: reads and persists
- Provider checks → Provider interface: uses interface
tools/rlm/provider_check.py - Provider checks → HTTP helpers: sends checks
- HTTP helpers → LLM endpoint: sends HTTP request
tools/rlm/providers/http_utils.py