Docs
AI Docs
Crystal Agents
Delivery Plan

Delivery Plan

This plan assumes two-week sprints and a small team focused on getting the first useful version into production quickly.

Current Status

  • Week 1: Done
  • Week 2: Done
  • Week 3: Done
  • Week 4: Done
  • Week 5: Done
  • Week 6: Done
  • Week 7: Partial
  • Week 8: Partial
  • Week 9: Done

Milestone View

  • Milestone 0, Project Skeleton: Done
  • Milestone 1, Core Orchestration MVP: Done
  • Milestone 2, Approval and Safety Controls: Done
  • Milestone 3, Operational Expansion: Done
  • Milestone 4, Visibility and Operator Experience: Done
  • Milestone 5, Smarter Routing and Memory: Not started
  • Milestone 6, App-Managed Scheduler and Script Automation: Done

Release Status

  • Current target release: v0.3
  • Scope: orchestrator, workflows, approval gates, dashboard shell, reports, operational workflows, Telegram notifications, model config, and the app-managed scheduler
  • Ready for: internal use and iterative expansion

Next Up

  1. Add smarter memory and routing only after the scheduler is stable.
  2. Expand scheduler dashboards and filters if the job count grows.
  3. Add stronger model routing and policy-driven selection per task type.

Week 1

Goal: create the repo skeleton and execution baseline.

Work

  • finalize the package layout
  • define task, run, agent, workflow, and tool data models
  • add Docker and Docker Compose
  • add a CLI entrypoint
  • set up logging and run artifact folders

Output

  • a runnable placeholder project
  • a predictable file structure for runs and logs

Week 2

Goal: establish the minimal control plane.

Work

  • add the orchestrator shell
  • add the workflow runner
  • add base agent and tool abstractions
  • wire the CLI to start a simple run

Output

  • a task can enter the system and move through a stub workflow

Week 3

Goal: ship the first useful workflow.

Work

  • implement the CEO, Developer, Tester, Research, and Content stubs
  • add repo read/write tools
  • add test runner and web search tool shells
  • implement feature spec generation
  • implement content drafting

Output

  • two workflows complete end to end and write outputs to run folders

Week 4

Goal: make the MVP reviewable and testable.

Work

  • add structured step logs
  • add workflow status tracking
  • add output formatting rules
  • add basic tests for workflow execution

Output

  • workflow execution is visible and repeatable

Week 5

Goal: lock down safety.

Work

  • add role-based tool permissions
  • add approval gates for risky actions
  • add audit trail records
  • add tests for denied tool calls

Output

  • unsafe actions are blocked unless approved

Week 6

Goal: validate the MVP in a realistic scenario.

Work

  • run the feature-spec workflow on a real request
  • run the content workflow on a real request
  • fix any broken logging, routing, or artifact creation
  • refine output templates

Output

  • the system can support a full request flow with human review

Week 7

Goal: expand into operations and growth tasks.

Work

  • add Reviewer, DevOps, Ads Manager, Sales Assistant, SEO, and Analytics agent stubs
  • add lead follow-up and campaign workflows
  • add bug triage and release preparation workflows

Output

  • the system covers more than development and content

Week 8

Goal: improve operator usability and prepare for the next release.

Work

  • create a dashboard shell
  • add run history browsing
  • add approval queue views
  • add reporting summaries

Output

  • users can inspect and approve work without digging through raw files

Release Gate

The first release is ready when:

  • the CLI can start and track runs
  • at least two workflows are stable
  • tool permissions are enforced
  • approvals are required for sensitive actions
  • logs and run artifacts are readable

Status Notes

  • The MVP control plane is implemented and working.
  • The approval flow and dashboard shell are implemented.
  • The operational expansion is implemented, including reusable templates and shared output formats.
  • The app-managed scheduler is implemented and uses allowlisted bash scripts, approval gates, and history logging.
  • The repo has been cleaned up for tagged v0.1 and v0.2 releases and is now moving toward v0.3.

Week 9

Goal: ship the app-managed scheduler for allowed bash jobs.

Work

  • add scheduler config models and cron parsing
  • add scheduler CLI commands for listing and running due jobs
  • run allowlisted bash scripts from the VPS on schedule
  • integrate approval gating and history logging for scheduled jobs
  • surface scheduler jobs and history in the dashboard

Output

  • the app can manage scheduled scripts without a separate agent
  • scheduled jobs are visible, auditable, and approval-aware
DashboardDeploy Walkthrough