aaix/laravel-tall-architect

Quick Start

composer require aaix/laravel-tall-architect --dev
php artisan boost:install

Boost lists third-party packages during its install — tick aaix/laravel-tall-architect there. Where Boost is already set up, add the package to the existing selection instead:

php artisan boost:update --discover

That is the whole setup. All seven guidelines are now part of every agent file; php artisan tall-architect:status shows the result.

What ships

Agent rules rot the moment they are copied. Here they stay a composer dependency, so a correction rolls out everywhere instead of into one repository at a time.

Choosing the guidelines

All seven are on by default, and each has its own flag:

TALL_ARCHITECT_NONTECH_USER=false
TALL_ARCHITECT_FEATURE_BOOK=false

Run php artisan boost:update afterwards to recompose the agent files. The flags control the always-on text; they do not remove the separately shipped skills.

To replace the shipped set with a project's own, point TALL_ARCHITECT_PATH at a directory holding files of the same names.

Feature Book and audits

The Feature Book preserves agreed user intent independently of the implementation, one chapter per feature. Chapters live at .ai/project/features/<domain>/<feature>.md, grouped by business capability, with stable feature IDs, an immutable creation date, and inline YAML tags such as tags: [billing, payments]. Every chapter uses three plain-language sections: What is it for?, How does it work?, and Examples. Feature-specific subheadings are welcome; rule IDs and code maps are not required. Technical implementation choices remain open unless explicitly required or necessary for the agreed behavior. Relevant technical details remain part of the chapter. A compact index.md provides navigation. Plans track unfinished work; the Feature Book states the agreed behavior.

The feature-book guideline tells agents to write chapters as part of the work, without offering or asking, to read relevant chapters before making changes and update affected passages with authorized behavior changes. Refactors do not require rewriting the specification. Code observations alone never establish requirements. Agents may suggest an audit after substantial changes or discovered contradictions; elapsed time alone does not trigger a reminder or an automatic audit.

Example requests:

• “Use feature-book to write the booking cancellation chapter from our agreed requirements. Clarify behavior inferred only from code before including it.”
• “Use feature-book-audit to comprehensively check billing against its chapters.”
• “Use feature-book-audit to review the whole Feature Book for up to two hours and save progress for resuming.”

Audits write coverage, evidence, findings, and continuation state to .ai/project/audits/<YYYY-MM-DD>_<scope>_<unique-run-id>/report.md. They distinguish static inspection from executed tests, extract checks from the chapters without imposing technical choices, preserve unresolved and pending checks, and do not repair code or rewrite requirements. Long runs can resume after checking for changed inputs. Time/token limits depend on available measurement; the skill cannot determine remaining account quota or provide background scheduling itself.

Keeping projects current

composer update aaix/laravel-tall-architect
php artisan boost:update

Commands

License

MIT