Claude Code pluginv0.1.0MIT

replicants

A generic engineering loadout for human-in-the-loop work: role-based agents and task skills that bind themselves to whatever repository Claude Code is running in.

01

How it adapts to your project

  1. Finds the root

    The git top level, or else the working directory.

  2. Reads CLAUDE.md

    Your conventions and prohibitions are the authoritative source.

  3. Probes the repo

    Stack, test runner, lint, build and deploy — a script that runs nothing counts as absent.

No agent invents a command, path, port, service or credential it did not find.

02

Agents

Invoke with @agent-replicants:<name>, or let Claude delegate.

AgentRole
codieImplements, refactors and debugs to Coding Excellence standards; leaves independent verification to qa
analystRead-only investigation: requirements, assumptions, alternatives, “is X built and where”
analyst-workerRead-only slice worker for analyst
qaVerifies criteria by running the project’s real tests; reports evidence per criterion
qa_workerReport-only test-slice worker for qa
architectDesigns and writes ADRs and design documents; read-only on source code
deployContainers, CI/CD and infrastructure-as-code for what the project already uses; asks before anything destructive

Agents use explicit built-in tool allowlists and no MCP tools. Read-only agents with shell access are instructed, not technically prevented — use Claude Code permissions for hard limits.

03

Skills

SkillWhat it does
/replicants:qaRepeatable QA pass: derive criteria, run the tests, report pass or fail with evidence
/replicants:ceApply Coding Excellence standards while writing or reviewing code
/replicants:confirm_denyConfirm, deny with the correction, or mark unresolved every assumption in your prompt
/replicants:cite_sourceBack every claim with a fresh lookup and a citation
/replicants:qrAsk clarifying questions with recommendations before implementing
/replicants:dontRead-only turn: advice only, with an approval gate before any change
/replicants:verbA fully spelled-out, plain-language answer for this turn
04

Install

claude plugin marketplace add rational-agents-ai/rational_agents_plugins --scope projectclaude plugin install replicants@rational-agents --scope project

Run from the repository you want the plugins in. Recorded in its .claude/settings.json, so everyone who works there gets the same setup.