Sitelayer Engineering Skill

Use this skill for ordinary engineering work in this repo.

Before Work

• Read the root agent doc (CLAUDE.md or AGENTS.md) when present.
• Read CONTEXT.md for domain language when present.
• Read relevant ADRs under docs/adr/.
• Read docs/agents/issue-tracker.md before creating or updating work items.

Debugging

• Build a deterministic feedback loop before hypothesizing: targeted test, CLI fixture, API script, trace replay, or a small harness.
• Reproduce the exact user-visible failure, then write falsifiable hypotheses.
• Instrument one hypothesis at a time and remove temporary logs before finishing.
• Turn the minimized repro into a regression test at the real seam when a good seam exists.

Implementation

• Prefer vertical slices: one observable behavior, one test, one implementation step, one verification loop.
• Test through public interfaces and runtime contracts, not private helper shape.
• Keep scope boundaries explicit when touching shared infrastructure or user-facing workflows.

Architecture

• Prefer deep modules: small interface, meaningful implementation behind it.
• Use the deletion test. If removing a module only removes indirection, it was shallow.
• The interface is the test surface. If the only useful test must reach through the interface, the module shape is suspect.

Task Work

• Treat the tracker documented in docs/agents/issue-tracker.md as the issue tracker for imported skills.
• Split plans into end-to-end tasks, not layer-only tasks.
• A task ready for an autonomous worker needs current behavior, desired behavior, acceptance criteria, scope boundaries, and verification commands.