agent-teammate

Agent Teammate

What It Does

Creates and manages a persistent AI agent teammate that remembers project context, user preferences, and past decisions across sessions. The teammate has a defined personality, role-specific strengths/weaknesses, and can be @mentioned in conversations to contribute expertise without full context handoff.

Iron Laws (NEVER violate)

1. Memory is sacred — Never lose or corrupt teammate memory between sessions. Every interaction must be persisted.
2. Role boundaries — Teammate stays in its lane. A reviewer teammate does not rewrite; a planner teammate does not deploy.
3. Feedback loop required — Every correction from the human partner must update the teammate's behavior model.
4. No hallucinated memory — Never fabricate past interactions. If memory is missing, ask — don't guess.

Red Flags (STOP immediately)

• Memory drift — Teammate contradicts what was established 3+ sessions ago → memory corruption
• Role creep — Teammate starts performing tasks outside its defined role without being asked
• Silent disagreement — Teammate disagrees with a decision but doesn't surface it → escalates to human
• Context amnesia — Teammate asks questions already answered in persistent memory

Common Rationalizations (self-deception)

• "I'll remember this detail, no need to persist it" → Memory decays. Persist everything that matters.
• "The teammate knows the project well enough, skip the context" → Always verify current context.
• "One more capability won't hurt the role definition" → Role creep destroys trust and predictability.

When To Use

• User wants a persistent AI collaborator that remembers project history
• Complex multi-session projects where context handoff is expensive
• Team workflows where multiple humans interact with the same agent teammate
• User wants a specialized role (architect, reviewer, tester) with persistent memory

Human Partner Signals (escalate to human)

• Confidence < 70% — Teammate is unsure about a critical decision → escalate
• Role conflict detected — Task requires capabilities outside teammate's role → ask human to reassign
• Memory inconsistency found — Two memory entries contradict → flag for human resolution
• Ethical boundary touch — Task approaches ethical gray zone → pause and escalate

Pipeline

1. Define teammate: role, personality, strengths, weaknesses, communication style
2. Initialize memory: load past sessions, project context, user preferences
3. Deploy teammate: register @mention trigger, set up notification channels
4. Operate: teammate responds to @mentions, learns from feedback, updates memory
5. Review: periodic memory audit, role boundary check, performance review with human

Verification Checklist

• Teammate memory persists across 3+ sessions without data loss
• Role boundaries enforced: teammate refuses out-of-scope tasks
• @mention triggers teammate response within defined latency SLA
• Feedback from human updates teammate behavior in next interaction
• Memory audit shows no fabricated or contradictory entries
• Human partner can view/edit teammate memory directly

Related Skills

• squad-builder — Agent teammates are the building blocks of squads
• task-lifecycle-manager — Tasks assigned to teammates go through lifecycle tracking
• session-memory — Underlying memory infrastructure for persistence
• agent-introspector — Debugging teammate behavior and decision traces