ai-agent-rules-generator task.instructions.md

Copilot Global Execution Rules

These rules apply to ALL tasks without exception. They are BLOCKING — you may not skip any step.

---

BLOCKING: Task Lifecycle (execute for EVERY task)

This is a gate-checked pipeline. Each step MUST complete before the next begins. Skipping any step is a rule violation.

Step 1 — BEFORE writing any code

• Add unchecked checkbox item(s) in docs/task/todo.md
• If task is multi-step (≥3 steps), create docs/task/planning/<task-slug>.md and reference it in the todo item
• For non-trivial requests: write a brief plan before implementation

STOP: Do NOT write implementation code until Step 1 is complete.

Step 2 — During implementation

• Keep progress status accurate in docs/task/todo.md (check boxes as sub-tasks complete)
• If new information invalidates the plan, STOP and re-plan immediately

Step 3 — After implementation, BEFORE marking complete

• Verify the change works (build, dev server, browser, endpoints — whatever applies)
• For UI changes: take screenshot to confirm rendering
• For API changes: test endpoint responses
• For database changes: verify migrations and data integrity

STOP: Do NOT mark the task complete until verification passes.

Step 4 — Log Observations (BLOCKING GATE)

• Follow .github/instructions/agent-observations.instructions.md
• Review everything you touched, read, or discovered during this task
• Append entries to docs/agent-observations/ logs: critical.md, recommendations.md, anomalies.md
• If no observations: explicitly state "Observations: none."

STOP: This gate MUST complete before Step 5. Silence is a violation.

Step 5 — Completion (all three are required)

• Remove the finished item from docs/task/todo.md
• Append a row to today's daily task log docs/task/logs/YYYY-MM-DD.md (see format below)
• Update relevant docs/logic/ documentation if the change touches plugin behavior