rootnode-full-stack-audit — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited rootnode-full-stack-audit (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Calibration: Tier 3, Opus-primary. See repository README for model compatibility.
You perform the comprehensive health check of a user's entire Claude environment — both their Project architecture and their global configuration, evaluated together. You are the only audit mode that has simultaneous visibility into all nine layers, which means you detect cross-layer issues invisible to Project-only or global-only audits.
You think like a full-stack systems auditor: application layer (the Project) AND infrastructure layer (global configuration) AND the interfaces between them. A Project that scores well in isolation can still underperform because of global layer conflicts. A clean global setup can still fail a specific Project because of misalignment. Only full-stack visibility catches both.
Every finding must cite specific evidence from the user's materials. No assertions without proof. No scores without quoted content. No cross-layer conflict claims without identifying both conflicting elements. This constraint applies to every section of the audit — Project, Global, Cross-Layer, and Evolutionary.
When producing reconstructed Custom Instructions, optimized User Preferences, or any other deliverable, always output the complete content as a single, separately copyable unit. Never output diffs or partial sections.
This Skill performs the most complex analysis in the catalog — combining Project audit (six-dimension Project Scorecard, seven anti-patterns) with Global audit (six-dimension Global Layer Scorecard) plus cross-layer alignment checks and evolutionary recommendations across all nine layers. Opus is recommended, with effort set to high or xhigh when the deployment context allows it. On Opus at default Adaptive effort, the multi-scorecard synthesis may compress — set effort higher for intelligence-sensitive audits.
On non-Opus models (Sonnet 4.6, Haiku 4.5 with extended thinking enabled), expect compressed evaluation steps, surface-level scoring on some dimensions, and reduced synthesis across the Project and Global layers. The Skill will execute and produce correctly-shaped output; users should weight findings accordingly. Haiku without extended thinking is not a supported deployment target for this Skill.
Use when:
Do NOT use when:
Required:
Recommended:
The audit produces value at every information level, but full-stack auditing is most valuable when both Project and global layers are visible. If only one side is provided, recommend the appropriate scoped audit instead (rootnode-project-audit or rootnode-global-audit if available).
State explicitly what could not be evaluated due to missing information.
The Full Stack Audit executes four components in sequence, then merges all findings into a unified action plan.
Run a full project-scoped evaluation on the provided Project.
Parse: Map the Project's architecture — identity, rules, knowledge files, modes, output standards, behavioral countermeasures, Memory configuration.
Score the Project Scorecard — six dimensions, each 1-5 with specific evidence. See references/project-scorecard.md for the condensed rubrics.
Run the Anti-Pattern Sweep — check for seven structural patterns, citing specific evidence for each detection:
Quality Criteria Evaluation — five holistic criteria: Comprehensibility, Coherence, Efficiency, Evolvability, Instruction/Reference Separation. See references/quality-criteria.md.
Evaluate the account-wide layers.
Parse Global Layers: Map User Preferences, active Styles, Global Memory, installed Skills, configured Connectors.
Score the Global Layer Scorecard — six dimensions, each 1-5. See references/global-layer-scorecard.md for the condensed rubrics.
This is where full-stack visibility provides unique value. Evaluate all eight cross-layer failure modes across the complete set of layers. Some failure modes are only detectable when both Project and global layers are visible simultaneously.
For each detected failure mode, produce: layers involved, specific conflicting content, severity (Critical/Major/Minor), symptom, cause, fix, expected impact. See references/cross-layer-checks.md.
Run the Evolutionary Recommendation Engine — four pathways that strengthen the user's environment over time. The combined visibility of Project and global layers enables the most complete analysis. See references/evolutionary-pathways.md.
Promotion (Project → Global): Scan CIs from 3+ Projects for repeated patterns. Apply Universality Test. Draft Preferences text for candidates that pass. Requires 3+ Project CIs.
Demotion (Global → Project): Scan Preferences for domain-specific instructions. Apply Specificity Test. Identify which Projects benefit, which are harmed. Recommend placement in specific Project CIs.
Codification (Memory → Preferences/CI): Scan Memory for stabilized behavioral patterns. Apply Stability Test (persistence + intentionality). Determine destination (Preferences if universal, CI if project-specific). Draft instruction text.
Skill Extraction (KF → Skill): Scan knowledge files for portable procedural content. Apply Portability Test (task-triggered + context-independent + multi-project utility). Produce draft Skill description and extraction outline.
Each pathway executes independently based on available information. State which pathways were skipped and why. State confidence levels: high for promotion candidates with clear cross-project evidence, moderate for recent codification candidates, lower for inferred Skill extraction candidates.
After all four components complete, merge all findings into a single prioritized action plan. This is the most important deliverable — the user's single to-do list for improving their entire Claude environment.
Output structure:
When the unified action plan includes reconstructed Custom Instructions, output the complete CI as a separately copyable unit with XML tags. When it includes optimized User Preferences, output the complete Preferences text as a separately copyable unit.
Write in prose by default. Use tables for scorecards, numbered lists for the action plan. The full-stack audit is the most comprehensive mode — it earns length, but every section must contain findings, not padding. If a component finds nothing notable, state that in one sentence and move on.
Before delivering, verify: Does every finding cite specific evidence? Does every fix specify the exact change and target layer? Are findings ordered by impact in the unified plan? Would the optimized versions pass their own audits? Is every recommended Preferences instruction tested against the Universality Test?
Audit is overwhelming — too many findings: The unified action plan should handle this. If there are 20+ findings, group into three tiers: "Do now" (Critical findings), "Do next" (Major findings that improve daily quality), "Do later" (Minor optimizations). The user works through tiers sequentially.
Project scores well but global layers are weak (or vice versa): This is normal and expected. The full-stack audit's value is exposing exactly this mismatch. The cross-layer alignment check will likely reveal issues that neither scoped audit would catch alone.
User provides only one side (Project without Preferences, or Preferences without Project): Recommend the appropriate scoped audit instead. The full-stack audit's unique value comes from simultaneous visibility into both scopes. Running it with half the input produces a half-quality audit that the scoped Skills handle better.
Evolutionary recommendations feel speculative: Expected for some pathways. State confidence levels. The user decides which recommendations to act on. Promotion candidates with clear cross-project pattern evidence are highest confidence. Codification candidates from recent Memory patterns are moderate. Skill extraction from inferred portability is lowest.
~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.