Playbook Laravel Package

Product Decisions This Supports

• Standardizing AI-assisted engineering across Laravel repos. v0.6.0 ships 11 role-specific Claude Code task agents (architect, auditor, debugger, pr-reviewer, qa-tester, data-backfill, etc.) plus a routing rule, turning ad-hoc prompting into a repeatable team capability. This supports a “standardize the workflow” decision rather than letting each engineer configure their own agents.
• Developer-productivity roadmap and cycle-time bets. The package targets repeatable, low-judgment work: PR review and fix-up, dependency updates, QA testing, debugging, architecture review, and data tasks. A PM can tie adoption to measurable outcomes like PR turnaround, dependency lag, QA coverage, and onboarding time.
• Build vs. buy for internal AI tooling. Adopting jiannius/playbook avoids maintaining an internal agent library and routing rules. config/playbook.php provides per-agent opt-outs (agents.except) and a global off switch (agents.enabled), which limits lock-in and allows a staged rollout.
• Data-access governance. The changed policy allows data investigation and backfills against staging only when the requester explicitly grants it, with production still off-limits. This supports a policy decision to permit AI-assisted data work with explicit human approval, rather than banning it outright or leaving it undefined.
• CI-enforced reproducibility. playbook:check now verifies the agent folder and routing rule, exits 1 until boost:update is run, and exits 2 if the directories are gitignored or if an agent name collides with a team-owned agent. This supports treating agent configuration as reviewed, committed infrastructure rather than local workspace state.
• Team onboarding and consistency. Because agents and rules are committed to the repo, new engineers inherit the same Claude Code setup. The cost is a required restart after pulling, and qa-tester needs a browser tool (Playwright) in Claude Code.

When to Consider This Package

Consider adoption when:

• The team uses Claude Code as its AI coding assistant. The routing rule is deliberately Claude-only because the agents’ model and effort settings are Claude settings; this is not a vendor-neutral agent framework.
• The codebase is Laravel/PHP and uses artisan, since installation and checks run through boost:update, boost:install, and playbook:check.
• You want repeatable, role-based agents for PR review, QA, dependency updates, debugging, architecture review, and controlled data tasks, rather than individual prompt collections.
• You are willing to commit generated agent files (.claude/agents/, .claude/rules/, .claude/skills/, AGENTS.md) and keep them updated via boost:update, ideally in post-update-cmd.
• You need guardrails around data work: staging requires explicit requester approval, production is off-limits, and @claude replies cannot start task agents via --disallowedTools "Agent".
• You want a CI gate on agent presence and configuration hygiene, including protection against gitignored agent folders and name collisions.

Look elsewhere when:

• The team uses GitHub Copilot, Cursor, or a multi-vendor AI stack and needs equivalent behavior across tools.
• The team does not want AI agents or generated configuration committed to repositories.
• You need a general-purpose agent orchestration framework rather than Laravel-specific workflow scaffolding.
• The team cannot absorb a coordinated upgrade: v0.6.0 is breaking, and playbook:check will fail until boost:update is run and files are committed.

How to Pitch It (Stakeholders)

For executives: This is a low-cost, open-source Laravel package that standardizes how our engineers use Claude Code. It installs a set of role-specific agents for common engineering work—PR review, QA, dependency updates, debugging, architecture review, and controlled data tasks—so AI assistance becomes a repeatable team capability rather than individual experimentation. It includes explicit data-safety controls: staging data work requires the requester’s approval, and production remains off-limits. Adoption is reversible per agent via configuration.

For engineering leadership: v0.6.0 is a breaking release. playbook:check now verifies that agent files and routing rules are present, exits 1 until boost:update is run, and exits 2 if the folders are gitignored or if a team-owned agent name collides with a packaged agent. The upgrade requires a coordinated commit of composer.*, AGENTS.md, .claude/skills/, .claude/agents/, and .claude/rules/. Teammates must restart Claude Code after pulling so the new agents folder is picked up, and qa-tester requires Playwright. The payoff is a committed, reviewable, CI-enforced AI workflow with per-agent opt-outs.

For the delivery team: Treat this as infrastructure, not a toy. Run php artisan boost:update, then php artisan playbook:check, and expect exit 0 before merging. Use config/playbook.php to opt out of specific agents or all agents, and resolve name clashes deliberately—either opt out of the packaged agent or delete the team-owned duplicate. Do not gitignore the playbook folders; the check treats that as an error by design.