Use this checklist after installing into your real ~/.copilot profile.

Runtime checks

Copilot CLI

  • /env shows the goal-system extension or plugin surface.
  • /skills info goal shows the goal skill.
  • Starting /goal <real task> creates a persisted goal instead of only acknowledging the skill.
  • goal_system_status shows the same goal ID after a new prompt in the same session.
  • A subagent receives main-session-only boundary text and cannot use goal tools.
  • After five non-goal tool calls without goal_system_update, a sixth non-goal tool is denied.
  • Trying to stop with an open goal is blocked by the hook.
  • goal_system_close refuses complete when required evidence is missing.
  • goal_system_close accepts complete only after inspection, resolution, verification, coverage, no remaining work, no blockers, and completion audit are recorded.

VS Code Copilot Chat

  • MCP: List Servers shows goalSystem.
  • MCP: Reset Cached Tools refreshes the goal_system_* tools after install or update.
  • The Goal System custom agent appears in Copilot Chat.
  • The GitHub Copilot Chat Hooks output channel shows goal hooks running.
  • Starting /goal <real task> through the Goal System custom agent creates a persisted goal with the injected sessionId and cwd.
  • Asking to continue a single same-directory active goal in a fresh VS Code Chat session loads that persisted goal instead of starting from memory.
  • Stop blocks while an open goal remains active.
  • SubagentStart gives subagents the boundary text without full goal details.

Fixture prompt

Run this inside Copilot after install:
@tests/prompts/goal-system-reliability-e2e.md
The fixture project has a real failing test. The live session should inspect it, run the test, fix it, rerun it, update the persisted goal, and close only after proof exists.

Known limits

Local tests cannot prove a specific Copilot CLI or VS Code build invokes every documented hook. Verify /env, MCP server registration, hook output, and one real /goal session after installing.