ragbaz-frog

Path: ragbaz-frog/

Frog is the workspace coordination CLI for /data/src. It manages task slices, locks, repos, units, files, and event logs through a SQLite database (AGENTS.db), providing both a human-oriented terminal interface and an MCP server for agent consumption.

Architecture

                ┌─────────────┐
                │  CLI (argparse)│
                │  frog <cmd>  │
                │  <subcmd>    │
                └──────┬──────┘
                       │
          ┌────────────┼────────────┐
          ▼            ▼            ▼
   ┌──────────┐ ┌──────────┐ ┌──────────┐
   │ Service  │ │ Service  │ │ Service  │
   │  Layer   │ │  Layer   │ │  Layer   │
   └────┬─────┘ └────┬─────┘ └────┬─────┘
        │            │            │
        └────────────┼────────────┘
                     ▼
              ┌──────────┐
              │  Store   │
              │ (SQLite) │
              │AGENTS.db │
              └──────────┘

Three layers:

1. Store Layer (ragbazfrog/store/)

SQLite database layer using aiosqlite. Manages:

• Schema migrations — version-tracked SQL migration files
• Tables: tasks, repos, locks, event_log, units, files, peers, box_identity, task_dependencies, task_assignments, task_tags, task_status_history, event_hooks, event_mirror, repo_aliases, repo_deps, repo_artifacts, repo_targets, target_runs, schema_migrations

2. Service Layer (ragbazfrog/services/)

Domain logic organized by concern:

• repo — repo registration, aliases, dependencies, units, targets, artifacts
• task — task CRUD, dependencies, assignments, tags, status history
• lock — lock acquire/release, audit, reap (stale recovery), list
• sync — pull workspace state from external sources
• box — box identity management for cross-box federation
• peer — peer join and communication for multi-box setups
• mcp — MCP tool definitions surfaced by the MCP server
• status — workspace summary (repos, files, tasks, locks)
• log — event log queries, follow/tail, why/blame
• unit — unit discovery and listing

3. CLI Layer (ragbazfrog/cli/)

Argparse-based command tree. Strict grammar: frog <command> <subcommand> [args].

frog db migrate
frog new NAME                    # scaffold a repo/draft
frog completion bash|fish
frog log [--follow]
frog repo list
frog repo info REPO
frog repo build [REPO]           # + test/lint/clean/check/verify/diff/
                                 #   status/scan/targets/doctor/artifacts/
                                 #   artifact-stale
frog repo task list --repo REPO
frog repo dep add DEPENDENT DEPENDENCY
frog repo dep list
frog repo affected [--since REF]
frog repo build --affected
frog lock acquire ...
frog lock audit
frog lock reap
frog lock list --status ...
frog task create ...
frog task next [--agent N]
frog unit discover [--repo REPO]
frog unit list
frog mcp tools
frog mcp serve
frog status
frog sync pull WS
frog sync list
frog log why SLUG
frog log blame FILE

MCP Integration

Frog also runs as an MCP server:

/data/src/ragbaz-frog/bin/frog-mcp

Exposes MCP tools for repo, unit, task, lock, log, status, and workspace operations. Agents should prefer frog MCP over guessing repo structure from raw directories.

Database Schema (AGENTS.db)

The SQLite database at /data/src/AGENTS.db contains 20+ tables:

Table	Purpose
tasks	Task slices with metadata (slug, description, status, priority, effort, impact)
task_dependencies	Directed dependency graph between tasks
task_assignments	Agent-to-task mapping
task_tags	Tag-based categorization
task_status_history	Immutable status change log
repos	Registered repo boundaries
repo_aliases	Alternative names for repos
repo_deps	Inter-repo dependency declarations
repo_artifacts	Build artifact tracking
repo_targets	Build target definitions
target_runs	Target execution history
units	Nested buildable/runnable subprojects
files	File classifications
locks	Coordination locks (agent, pid, repo, host, lease)
peers	Known peer boxes for federation
box_identity	Local box identity
event_log	Immutable coordination event log
event_hooks	Event-triggered hook definitions
event_mirror	Event mirroring to peers
schema_migrations	Migration version tracking

Key Concepts

Workspace

The source root (/data/src). Treated as a structured workspace, not an undifferentiated directory tree.

Repo

A tracked project boundary. Not every nested directory is a repo — prefer frog metadata over path name inference.

Unit

A nested buildable or runnable subproject inside a repo. Used for suites with multiple levels.

Lock

Coordination locks stored in AGENTS.db (not pid files). Support lease renewal and stale-lock recovery. Lock records include agent, pid, repo, host, start time, lease info.

Task

Descriptive task slugs stable enough for commit messages and cross-references. Tracked with metadata: why, what, ROI/impact/effort, dependencies, conflicts, status history.

Federation

Frog supports cross-box federation through box identity, peer joining, and event mirroring. Locks, tasks, and events can span multiple machines.

Implementation

• Language: Python
• Database: SQLite (aiosqlite)
• CLI: argparse
• MCP: MCP protocol server
• Service pattern: Store (data access) → Service (business logic) → CLI/MCP (presentation)
• Config: Box identity, peer list, event hooks configured through the store

Current Status

Active and well-tested (62+ tests). The CLI has gone through a major redesign (Phase I: strict grammar, Phase II: TUI/board parity, federation, MCP standards). 85+ completed tasks. Used as the coordination backbone for all multi-agent work under /data/src.

---

See Products Overview for the full portfolio. Frog runs alongside host services such as Traefik and Infisical.