Agents Workspace

Agents Workspace

**A lightweight agents and skills set to handle any kind of task!**

Designed to create a professional workflow with dynamic team assembly of specialized agents tailored to each task, parallel execution, optimized context windows, and a self-learning cycle.

Under the hood, its operating model brings together memory, subagent-driven development, spec-driven workflows, and agent loops grounded in empirical verification.

Installs [The Agency](https://github.com/msitarzewski/agency-agents), a complete AI-agent package covering software engineering, design, marketing, sales, finance, paid media, project management, and other task domains.

Harness Flow

The user-facing main agent is the **coordinator**, not the executor. It gathers lean workspace context, assembles the right team, delegates scoped work, reviews the results, and synthesizes the final response. The parallel branches below illustrate the default pattern for independent scopes; orchestration can use a single-agent or sequential pattern when the task or its dependencies call for it.

Review and verification are coordinator-controlled quality gates: when findings show that the done criteria are not met, the coordinator re-dispatches the affected subagent(s) with the findings, integrates the adjustments, and sends the result through review again. The cycle repeats under the normal status and escalation rules until the criteria pass; blocked or stuck work is stopped, escalated, or recomposed rather than retried without bound.

User request
     |
     v
+--------------------------------------+
| Main agent (coordinator)             |
| Read wiki/index.md for context       |
| Optional: use available knowledge    |
| tools (e.g. code/symbol/dependency   |
| graphs).                             |
+--------------------------------------+
     |
     +--> [when a durable behavior contract is needed]
     |       Load spec-builder; confirm spec workflow
     |
     v
+--------------