Built-in Agent: Aegis — Responsible AI Reviewer #1032

Summary

Add Aegis as Squad's third built-in agent alongside Scribe and Ralph. Aegis is a Responsible AI (RAI) reviewer that helps teams "start right, stay right" — reviewing work through an RAI lens for content safety, bias, terminology, and ethical patterns.

Like Scribe (memory) and Ralph (work monitoring), Aegis is a foundational capability that ensures every Squad team has RAI awareness from day one. You can't forget to add it — it's always there.

Philosophy

"Guardrail, not wall." Aegis helps fix issues, not just flag them. It's direct, practical, and empowering — never moralizing, never bureaucratic. Every finding includes WHAT's wrong, WHY it matters, and HOW to fix it.

Why Built-In?

RAI concerns are foundational, not optional. Making Aegis built-in means:

• Every team has RAI awareness from day one
• No risk of "we forgot to add a reviewer"
• Normalizes RAI as standard practice, not a special case
• Aligns with the "start right, stay right" philosophy

Proposed Design

Identity

• Name: Aegis (always "Aegis" — exempt from casting, like Scribe and Ralph)
• Emoji: 🛡️
• Roster entry: | Aegis | RAI Reviewer | .squad/agents/aegis/charter.md | 🛡️ RAI |

Traffic Light Verdict Model

When Aegis issues a Red, strict lockout semantics apply: original author is locked out, Aegis recommends a fix agent, and provides real-time guidance during revision (pair mode).

Activation Modes

Decision Authority

Aegis is advisory by default with escalation:

• Runs in background mode by default (like Scribe) — non-blocking
• Only escalates to blocking gate when a 🔴 Critical issue is found
• Fast-path bypass for low-risk changes (docs-only, test files, dependency bumps)

Files Owned

.squad/
├── agents/
│   └── aegis/
│       ├── charter.md        # Identity, responsibilities, voice
│       └── history.md        # Learnings across sessions
└── rai/
    ├── policy.md             # Canonical RAI policy (terms, anti-patterns, taxonomy)
    └── audit-trail.md        # Evidence log (append-only)

Check Categories (Phase 1 — High-Signal Only)

Starting narrow with checks that have clear, actionable fixes:

Code Review:

• 🔴 Hardcoded credentials / API keys / secrets
• 🔴 SQL injection, command injection, path traversal
• 🟡 PII exposure in logs or responses
• 🟡 Bias indicators in algorithms (demographic features, proxy attributes)
• 🟡 Missing rate limiting on user-facing endpoints

Content Review:

• 🔴 Harmful content patterns (hate speech, violence, self-harm)
• 🔴 Deceptive content (ungrounded claims, hallucinated citations)
• 🟡 Exclusionary language (gendered, ableist, culturally assumptive terms)

Prompt/Charter Review:

• 🔴 Instructions that bypass safety guidelines
• 🟡 Insufficient grounding for factual claims
• 🟡 Privacy/security risks in prompt design

Decision Review:

• 🟡 Unintended consequences (privacy regressions, accessibility impacts)
• 🟡 Stakeholder exclusion in design decisions

Project Type Awareness

Aegis calibrates based on what you're building:

Non-AI projects get minimal mode — high-signal checks without advisory noise.

Risk Mitigations

These were identified through team review and are required for launch:

Opt-Out Model (Tiered, Not Binary)

• Can't disable 🔴 Critical checks (credential leaks, hate speech)
• Can disable 🟡 Advisory checks with justification
• Temporary opt-down supported (auto re-enables)

Integration Points in squad.agent.md

• Add Aegis section after Ralph (identity, triggers, ceremonies, files owned)
• Add casting exemption
• Add to Init Mode roster proposal
• Add 🛡️ to role emoji table
• Add Aegis routing signals to routing table
• Add .squad/rai/* to Source of Truth hierarchy
• Update .gitattributes template

New Files

• templates/aegis-charter.md
• templates/rai-policy.md (default RAI policy template)
• Ceremony definitions: Pre-Ship Review + RAI Fix Guidance
• Aegis spawn prompt template

Implementation Phases

Phase 1: Core

• Add Aegis to squad.agent.md governance
• Create charter template + default RAI policy
• Implement high-signal check suite only
• Background mode default

Phase 2: Ceremonies

• Pre-Ship Review auto-trigger
• RAI Fix Guidance ceremony (pair mode after rejection)
• Integration with existing Reviewer Rejection Protocol

Phase 3: Intelligence

• Project type detection and check suite calibration
• False positive tracking and auto-disable
• Learn from user dismissals

Phase 4: Expansion

• Expand check suite based on real-world usage data
• Community-contributed check patterns
• Context-aware severity (test files get lower thresholds)

Success Metrics (measure after 3 months)

Kill Criteria

Abort if after tuning:

• False positive rate >40%
• Average review time >10 seconds

• 50% of teams opt out within first month

• Zero critical issues found in first 100 reviews

Context

This came from conversations with RAI experts about making Responsible AI a first-class concern in Squad — not an afterthought. The core insight: RAI should be as automatic as memory (Scribe) and work monitoring (Ralph). Teams shouldn't have to remember to think about RAI — it should be baked in.

The name "Aegis" (protective shield) captures the philosophy: protection that enables, not restricts.

Filed by the Squad team. Three agents collaborated on this proposal: architecture design, RAI pattern extraction, and critical risk review.