root.node Skills
The architecture system for Claude.
27 Skills that diagnose, build, and optimize every architectural layer across chat-Projects, conversations, and Claude Code repos.
Quick Start
root.node ships as six plugins in one marketplace. Install them on your claude.ai account once and the Skills load in chat, Cowork, and Claude Code.
claude.ai (recommended). Open Customize > Plugins > Add > Add marketplace, enter drayline/rootnode-skills, and turn on Sync automatically. Then add the plugins you want. Claude Code signed in with the same account loads them as <plugin>@synced.
Claude Code with an API key.
claude plugin marketplace add drayline/rootnode-skills
claude plugin install rootnode-projects@rootnode
Repeat the install for each plugin you want. Auto-update is off by default for third-party marketplaces; run claude plugin update to pull a new version.
Manual. The latest release carries the six plugin zips. Upload one at Customize > Plugins > Add > Upload plugin. A manual upload does not update itself.
Install each plugin once — on your account or locally in Claude Code, not both — or Claude Code loads two copies.
| Plugin | Skills | For |
|---|
rootnode-projects | project-audit, global-audit, full-stack-audit, anti-pattern-detection, memory-optimization, context-budget, project-brief | Auditing and optimizing Claude Projects and the claude.ai global layer |
rootnode-code | cc-design, repo-hygiene, plus a migration guard and a bounded test-evidence Stop hook | Designing and maintaining Claude Code environments |
rootnode-prompts | prompt-compilation, prompt-validation, block-selection, identity-blocks, reasoning-blocks, output-blocks, and five domain packs | Compiling and scoring prompts |
rootnode-tuning | behavioral-tuning | Countermeasures for Claude's behavioral tendencies |
rootnode-governance | critic-gate, mode-router, handoff-trigger-check, session-handoff, profile-builder | Governing autonomous work |
rootnode-skill-builder | skill-builder | Building, reviewing, and evaluating Skills |
Upgrading from v5.x or earlier? v6.0 renamed every Skill: the rootnode- prefix moved from the Skill to its plugin, so rootnode-project-audit is now rootnode-projects:project-audit. After you add the plugins, turn off or delete the old rootnode-* Skills where you uploaded them, and remove any rootnode-* folders from ~/.claude/skills/. Otherwise both copies load.
Evidence status. rootnode-projects ships the catalog's first eval suite (plugins/rootnode-projects/evals/, 5 cases for claude plugin eval). Its first run showed no lift over the base model: every case scored the same with and without the plugin, because the graders did not test anything the base model misses. Stricter graders and suites for the other plugins come next. Until then, treat the plugins as unmeasured.
The rootnode-code hooks run in every repository where that plugin is enabled. Chat ignores hooks.
[!NOTE]
Activate any Skill with natural language — no commands, no special syntax. Every Skill also works standalone.
[!TIP]
Calibrated for Claude Opus 5.5 + Sonnet 5.5 (dual-primary), v5.0. Fable 5.1 integrated-aware; Haiku 5.5 subagent-aware; Opus 5 / Sonnet 5 fallback-graceful (Skills produce correct-shape output there when classifier-flagged requests fall back silently); Sonnet 4.6 legacy-graceful. Most Skills work fully across the current model landscape; a few produce best results on Opus 5.5 (at its medium default or high) or Sonnet 5.5 at high. See Model Compatibility for details.
Why Architecture Matters
The Claude ecosystem has a layered architecture. How content is distributed across those layers determines whether Claude performs at its potential or works against itself. The hard part isn't building a Project or wiring a Claude Code repo — it's seeing what's quietly working against you.
Architectural inefficiencies don't announce themselves. They surface as vague symptoms: output that drifts, instructions that get ignored intermittently, quality that degrades over long conversations, autonomous agents that escalate work they shouldn't or barrel through halts they should respect. These Skills surface those structural issues, triage them by impact, and produce targeted fixes.
The trap is that these symptoms look like behavior problems. They're not. A rule the user has rewritten five different ways still gets ignored because the layer it sits in is wrong, not because the rule is unclear. Cross-layer conflicts cause unpredictable output that no amount of prompt rewording will fix. Misplaced enforcement in a Claude Code environment becomes preference instead of guarantee. Structural problems require structural solutions.
These Skills treat the entire ecosystem as a unified system — scoring each layer, detecting cross-layer conflicts, diagnosing behavioral tendencies, gating handoffs to autonomous execution, and building Projects and Claude Code environments that are structurally sound from the start. The system is calibrated to how Claude actually processes context: the behavioral tendencies documented through extensive testing, the loading behavior of each layer, and the architectural patterns that produce reliable output across both surfaces.
The Architectural Layers
Two surfaces, layered architecture across each. Skills are mapped to the layers they impact. A Skill that builds knowledge files appears on the Knowledge Files layer because that's the layer it modifies; a Skill that audits five layers at once appears on each of them.
This is the beginning of the README. Read the rest at the original. Read the full README ↗
