Skip to content
TilloTech Docs

Agent Workflow

Tillo Core should be legible to coding agents without hiding Laravel-specific correctness behind generic advice. The harness has three layers:

  1. AGENTS.md / CLAUDE.md are generated by Boost and injected into agent context. Do not edit them directly.
  2. docs/ is the repository-local system of record.
  3. Laravel Boost and .ai/skills/ provide executable, task-specific guidance.

Operating Model

Agents should gather only the context needed for the current task, then verify the result with the smallest relevant checks. If a rule or workflow is repeatedly needed, encode it where agents can discover and reuse it.

Use this sequence for most tasks:

  1. Read docs/index.md and ARCHITECTURE.md when the task touches unfamiliar territory.
  2. Activate the relevant skill from .ai/skills/.
  3. Use Laravel Boost for Laravel framework behavior, Artisan commands, database inspection, tinker, browser logs, and version-specific documentation.
  4. Inspect sibling files before adding new structure.
  5. Make the smallest coherent change that satisfies the requested behavior.
  6. Run focused verification. When ui/ changed, run nix develop -c npm run format:all and nix develop -c npm run type:check before finalizing (see .ai/guidelines/frontend-verification.md).
  7. Update docs or skills only when the task changes durable project knowledge.

Laravel Boost Is Part Of The Harness

Laravel Boost is the Laravel-aware layer of this agent system. Do not replace Boost guidance with generic docs. Use Boost when:

  • changing routing, middleware, validation, resources, policies, queues, migrations, or tests;
  • confirming Laravel 12 behavior;
  • checking Artisan command options before generation;
  • inspecting schema or reading database state;
  • debugging with tinker or browser logs.

Project docs should explain where Core-specific decisions live. Boost should remain the source for framework-version detail.

Knowledge Capture

Move hidden or repeated knowledge into the repository:

  • Slack decisions that affect implementation become ADRs or module docs.
  • Repeated PR comments become docs, tests, or lint rules.
  • Debugging discoveries become focused troubleshooting entries.
  • Complex rollout plans become versioned plans or ADRs when they will matter later.

Avoid one-off documentation churn. Capture knowledge when it will improve future agent or teammate decisions.