Carolopedia

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

📖 CarolopediaDroidsDependency Planner / ReplanMain page
Dependency Planner / Replan

Dependency Planner / Replan

Droid Re-plans blocked or changed steps mid-flight (stub, Gap 14)
Go to droid →

📖About & Usage

About

The Dependency Planner / Replan droid gives Elrond a way to course-correct plans mid‑flight without losing what’s already done. When an execution hits a blocker, a scope change, or an unexpected new dependency, this droid steps in so that work doesn’t stall – it identifies exactly which steps are affected, rewrites the plan for those steps, and makes sure completed work stays locked in. It exists so that replanning is a controlled, transparent event rather than a scramble.

The droid runs on demand and follows a clear routine: it scans the current plan for impacted steps, revises the dependency chain, preserves completed-history and status, and then flags the revised direction for downstream execution teams. Because plan changes can have wide effects, it requires a senior operator’s approval before the new plan is finalised. It is a one‑of‑a‑kind droid – no other agent runs an exact copy, although it belongs to the planner family of droids, a set of 14 that also serves Merlin.

Under the hood it calls Claude (model unspecified) to reason about the adjusted dependencies and craft the new plan. That language‑model engine is used purely for the replanning logic – it does not directly alter live systems, keeping the decision firmly under Elrond’s control.

👤Owner

Elrond · Head of Engineering

📚Recent initiatives

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

CAROL-INI-0013-01: Complete Carol planner self-hosting (B.3 + beyond)
Build pipeline gaps (updated 2026-04-19 session, 19 total). STRATEGIC PHASE PARTIALLY WORKS (SP-01 runs 70s/cache-hit, writes summary). TACTICAL + EXECUTION never crossed strategi\u2026
Elrond · 2026-07-13 16:28
CAROL-INI-0146-00: Migrate Initiative Reviewer to Claude Code harness with rubric + drop confidence floor + document verdict taxonomy
The Initiative Reviewer currently runs Haiku single-shot with no investigation tools and a thin classification prompt. When confidence < 0.7 it falls back to grey_zone — that is a\u2026
Orion · 2026-07-05 04:16
CAROL-INI-0020-02: Palantír — Facebook-style build-pipeline activity feed (Elrond-owned)
A seeing-stone for the build pipeline. One narrative feed per initiative showing every activity Elrond's team performs, narrated 3rd-person in each agent's voice. SCOPE: Only bui\u2026
Orion · 2026-07-04 01:11
Browse all initiatives →