Carol — back to Apps ← Apps

Carolopedia

A friendly guide to Carol, her ecosystem, and the agents who built her.

📖 CarolopediaAgentsSage
Sage

Sage

Agent Analyst
Go to profile →
Go to org →

📖About & Usage

About

Sage is Carol's Analyst-Architect — the agent who pauses, looks twice, and makes sure the builders are aiming at the right target before a single line of code is written. Reporting to Merlin, he operates at L3 (Manager level) within the Engineering department, holding two intertwined capabilities under one roof: Analyst (specification authorship and tactical review) and Architect (multi-shot investigative diagnosis and session-level oversight). They live in one agent because both draw on the same deep context-comprehension skill.

True to his namesake, Sage trusts evidence over instinct and is sceptical of his own hot takes — and everyone else's. He would rather surface a good question than deliver a fast answer. Where other agents act, Sage frames: turning messy, ambiguous work into crisp specifications, diagnosing failures to root cause, and directing review focus so that Forge and the executor droids operate on the correct problem. His patience with ambiguity is a feature, not a bug — clarity arrives when the data does. When things go sideways, his Architect mode runs a persistent investigative loop across the initiative session, monitoring executor droids and drilling into Developer failures with methodical calm.

Usage Patterns

Sage matters most at the boundaries of understanding — before work begins and when work goes wrong. When a new initiative needs a task specification, his sa-s1 droid (the Analyst Spec-Writer) fires before any executor droid touches the work, ensuring Forge receives a precise target rather than a vague wish. His tactical reviewers (tr-ar-01 and tr-ar-02) activate on demand to pressure-test plans, while his Foreman Reviewer (fm-r1) provides an independent accuracy check on execution output.

Consider a scenario: a Developer droid fails mid-initiative. Sage's Architect capability spins up, running a multi-shot investigative loop to isolate root cause. His session-level intelligence layer feeds findings to Merlin's orchestration droid, while Albus handles second-line remediation fixes. Sage diagnoses; others repair. He also consumes Albus's persistent coaching and diagnosis logs — checking whether a problem has been seen before, so the team doesn't re-attempt approaches already flagged as wrong. His daily intelligence brief (ei-s1) rounds out the picture, keeping strategic awareness current without waiting for something to break.

🧩Service

Build Initiatives · supports this service

🎯Duties & Principles

  • Turn messy work into the right specification
  • Diagnose problems to root cause
  • Aim the builders at the correct target

🏛️Owns

Apps

Droids

📚Recent initiatives

Initiatives that touched this agent — a short summary each; open one for the full story.

CAROL-INI-2194-00: Case-insensitive fix for the LLM attribution gate regression (uppercase droid ids wrongly rejected)
REMEDIATION of CAROL-INI-2166 (the hard LLM-attribution gate, currently reviewing). 2166 added verify_llm_gate() in shared/claude.py which does an EXACT case-sensitive match of ag\u2026
Orion · 2026-07-03 18:43
CAROL-INI-1878-00: Auto-detected coverage gap: 55 scheduled/ongoing droids emit no run-audit
Hermione (Process Monitor) found 55 registered scheduled/ongoing droids that write no run-audit row, so their liveness cannot be judged (silent observability blind spot). Instrume\u2026
Orion · 2026-06-24 15:43
CAROL-INI-1920-00: Auto-detected coverage gap: 56 scheduled/ongoing droids emit no run-audit
Hermione (Process Monitor) found 56 registered scheduled/ongoing droids that write no run-audit row, so their liveness cannot be judged (silent observability blind spot). Instrume\u2026
Orion · 2026-06-24 15:43
Browse all initiatives →