Claude Code plugins by Backchain. All plugins in this repository conform to the agentskills.io specification and are released as open source under Apache-2.0.
Team of Rivals
AI assistants tend to agree with you, even when your idea is... ill-advised. This skill set provides a panel of role-specific advisory agents that argue from focussed perspectives. Inspired by Lincoln's executive cabinet strategy: appoint your most challenging opponents to senior roles, rather than loyalists, so you receive direct and honest feedback from different perspectives. They will disagree with each other and they will disagree with you. The result is synthesized into a single honest recommendation designed to help identify blind spots and promote critical thinking.
| Agent | Perspective |
|---|---|
| Visionary | Expands possibilities, challenges constraints, identifies moonshot opportunities |
| Skeptic | Demands evidence, calculates failure probabilities, exposes hidden assumptions |
| Operator | Maps resources, identifies dependencies, grounds vision in execution reality |
| Ethicist | Evaluates stakeholder impact, assesses long-term consequences, applies moral frameworks |
Filesystem-backed agent memory
Three skills for explicit memory operations that complement Claude Code's native auto memory and /recap. consolidate audits knowledge directories and graduates stable feedback memories to permanent rules. briefing produces a cross-tool session-start orientation from your issue tracker, git history, and staleness signals. working manages an ephemeral .memory/ directory with an explicit checkpoint → promote → cleanup lifecycle.
| Skill | Purpose |
|---|---|
| consolidate | Audit knowledge dirs and auto memory; produce an editable cleanup plan; execute approved actions |
| briefing | Session-start briefing from issue tracker + git + working memory + staleness signals |
| working | Structured ephemeral state — todos, decisions, questions — with promotion to permanent locations |
Tool-agnostic: prompts the user for issue-tracker / ADR / docs locations rather than assuming a specific stack.
AI-slop content audit
Named for the Cynic who walked Athens in daylight with a lit lamp looking for an honest man. The audit skill (invoked as /diogenes:audit) forks a Sonnet subagent into a clean context, immune to upstream conversation bias, and judges whether a piece of writing reads as a competent human's work or as forwarded LLM output. It returns a verdict, an attributable findings table (quoted spans, named patterns, cited research), structural metrics, and a human rewrite of the weakest passage.
| Element | Purpose |
|---|---|
| audit | Six-category detection framework, required output format, and failure modes |
| diogenes | Senior-reviewer subagent (Sonnet, read-only tools plus WebFetch) that runs the audit |
Findings cite a fixed set of three peer-reviewed papers (Juzek & Ward 2025, Muñoz-Ortiz 2024, Reinhart et al. 2025); citations outside that set are forbidden.
/plugin marketplace add backchainai/backchain-plugins
/plugin install advisors@backchain-plugins
/plugin install engram@backchain-plugins
/plugin install diogenes@backchain-plugins
git clone https://github.com/backchainai/backchain-plugins.git
/plugin marketplace add ./backchain-plugins
/plugin install advisors@backchain-plugins
/plugin install engram@backchain-plugins
/plugin install diogenes@backchain-pluginsLicensed under the Apache License 2.0. Copyright (C) 2026 Backchain LLC.
Use, modify, and redistribute these plugins freely, including in commercial and proprietary work, under the terms of the license. See NOTICE for attribution that travels with redistribution.
Published by Backchain: Discover Where AI Works. Backchain builds practical agent tooling for Claude Code. Authored by Chris Krough (LinkedIn).
If these plugins are useful, install them, star the repo, and explore more at backchain.ai.
Contributions are welcome under Apache-2.0; contributors retain copyright on their work. No CLA and no DCO sign-off required. See CONTRIBUTING.md for the per-PR checklist.