remember

RustyLindquist's avatarfrom RustyLindquist

Reloads critical instructions when behavior degrades. Use when forgetting tools, not delegating, or missing safety rules. Refreshes orchestration mode, tool awareness, and safety constraints.

⭐0stars🔀0forks📁View on GitHub🕐Updated Jan 11, 2026
[Memory and Context Management]

When & Why to Use This Skill

The 'Remember' skill is a vital context-management utility designed to refresh an agent's core instructions, safety protocols, and tool awareness during extended sessions. It prevents behavioral drift and ensures the agent maintains its role as an orchestrator, effectively delegating tasks and adhering to safety constraints without losing session history.

Use Cases

  • Preventing behavioral degradation in long-running sessions by refreshing instructions every 30-45 minutes to maintain focus.
  • Correcting instances where the agent forgets to use specific tools like Supabase CLI or Playwright MCP for specialized tasks.
  • Re-establishing the orchestrator-delegate hierarchy when the agent starts performing implementation work directly instead of spawning specialized agents.
  • Reinforcing safety guardrails and non-destructive command policies when the agent's behavior becomes risky or deviates from established rules.
  • Quickly recovering the agent's operational state when a user notices the AI is missing context or ignoring previously established project rules.
nameremember
descriptionReloads critical instructions when behavior degrades. Use when forgetting tools, not delegating, or missing safety rules. Refreshes orchestration mode, tool awareness, and safety constraints.
allowed-toolsRead

Remember

Forcefully reload critical instructions without losing session state.

When to Use

  • Agent doing work directly instead of delegating
  • Forgetting tools (Supabase CLI, Playwright MCP)
  • Missing safety rules or not spawning agents
  • User says "you forgot" or "remember to..."
  • Every 30-45 min in long sessions

Immediate Actions

1. Confirm Safety

  • NO destructive DB commands (supabase db reset, DROP TABLE, TRUNCATE)
  • Use targeted SQL, createAdminClient(), incremental migrations instead
  • Inject safety rules into any spawned agents

2. Confirm Orchestration

  • I am the ORCHESTRATOR — I plan, coordinate, synthesize
  • I DELEGATE implementation to specialized agents
  • Frontend work → @frontend-agent
  • Backend work → @backend-agent
  • Code exploration → @research-agent
  • Testing → @test-agent

3. Confirm Tool Awareness

  • Supabase CLI — Use for ALL database questions
  • Playwright MCP — Use AFTER any UI changes
  • Agent spawning — Use for implementation work

4. Confirm Process

  • 2-gate flow: doc-informed plan → execute with doc updates
  • Context management: spawn agents, query lazily, checkpoint regularly

Output

## Instructions Refreshed

Safety: ACTIVE | Orchestration: ACTIVE | Tools: LOADED

Behavioral Check:
- [ ] Delegate frontend → @frontend-agent
- [ ] Delegate backend → @backend-agent
- [ ] Use Supabase CLI for database
- [ ] Use Playwright MCP for UI
- [ ] NO destructive commands

Ready to continue.