Phpstan Rules Laravel Package

Agents

This repository uses voku/agent-loop for governed coding work. Keep this always-on router small; detailed procedures live in the installed skills and CLI help. If host integration is uncertain, run tools/agent-loop/vendor/bin/agent-loop init host-status --format=json and obey its nextactionkind /...

claude-code

cursor

agent-guidance-maintenance

Maintain package-owned and host-owned agent skills, hooks, docs, sync targets, dogfood evidence, provenance capability mapping, and migration-safe validation.

claude-code

agent-learning

Capture reusable lessons about agent-loop workflow, validation, migration, evidence integrity, discipline dogfood, and package ownership before promoting durable guidance.

claude-code

agent-loop-code-review

Govern a read-only review of the complete raw diff, routing one dominant engineering lens and at most one focused handoff while preserving exact evidence and deterministic terminal state.

claude-code

agent-loop-dogfood

Evaluate agent guidance against real agent-* tasks with clean, comparable runs and observable artifact metrics instead of invented token savings.

claude-code

agent-loop-l2-context

Use agent-loop to compile, inspect, and govern task-scoped Recall/L2 context without duplicating agent-recall-compiler's prompt schema or treating generated artifacts as executed work.

claude-code

agent-loop-learning-boundary

Handle reusable knowledge that surfaces during a task, record honest Recall outcomes, close the governed Run learning decision, and keep findings separate from reviewed durable guidance.

claude-code

agent-loop-review-close

Review, verify, and close an agent-loop task safely after implementation, including code review, blind-spot review, truthful Recall outcomes, Learning validation, governed Run learning close-out, accepted-risk boundaries, and optional reflection.

claude-code

agent-loop-shipping-evidence

Prove merged, shipped, or released claims against an exact candidate SHA and exact integrated commit instead of branch names or historical PR state.

claude-code

agent-loop-simplify-audit

Run a bounded repo-wide complexity audit through the canonical code-review-simplicity lens, using agent-map only to choose and verify candidates.

claude-code

agent-loop-simplify-review

Run a read-only diff review through the canonical code-review-simplicity engineering lens, adding only agent-loop evidence and package-boundary context.

claude-code

agent-loop-surgical-edit

Apply an already-localized 1-2 file PHP change with the smallest verified diff. Use agent-map evidence, load coding-simplicity for implementation choices when installed, prefer deterministic agent-loop edit when possible, and escalate instead of silently widening scope.

claude-code

agent-loop-task-progress

Record bounded working memory during an agent-loop task, including decisions, checkpoints, validation evidence, scope changes, simplification ceilings, and blockers without copying raw evidence.

claude-code

agent-loop-task-start

Define durable task intent for a governed agent-loop task, then route startup through the lifecycle kernel instead of reproducing preparation or discovery choreography in host guidance.

claude-code

agent-loop-workflow

Operate the ordinary governed agent-loop path by routing through enter/finish and obeying the lifecycle kernel's structured next step instead of reproducing workflow policy in host prose.

claude-code

agent-recall-compiler-maintainer

Maintain voku/agent-recall-compiler package source, tests, docs, prompt output, manifest loading, scope selection, outcome logging, and local vendor syncs in consuming projects.

claude-code

agent-recall-consumer

Use voku/agent-recall-compiler to compile task-scoped Recall briefings, embed Recall through its typed PHP API, apply explicit operating-prompt recipes, review implementations, and record evidence-backed outcomes.

claude-code

Agent Loop Discipline

name: agent-loop-discipline description: Governed agent- orchestration: resumable state, map-first navigation, exact evidence, L2 gates, token-efficient agent I/O, review routing, and guidance changes.

claude-code

Agent Recall Consumer Review Evidence

When a PHP lifecycle host needs to inspect the current blind-spot review, use the Recall-owned read API instead of locating or parsing .blindspots.json itself:

claude-code

Agent Loop Investigate

name: agent-loop-investigate description: Locate PHP definitions, callers, tests, change sites, and evidence-backed temporal relationships with agent-map and bounded source reads. Read-only: report exact path/line/symbol evidence and do not propose or apply fixes.

claude-code

Agent Recall Consumer Embedding

When Recall is embedded by another PHP package, prefer the typed owner API:

claude-code