TAR Engine · Test · Audit · Report
A public guide of independent SKILL.md audits, plus a Playground that runs the same engine on whatever you paste in. Open source, BYOK for advanced layers.
Ten reader personas, sixty subcategories. Pick the lens that matches the work you do.
Backend, frontend, mobile, ops, security, testing
Data engineering, ML, MCP tools, prompts and RAG
Product management, growth, strategy, analytics
UI/UX, visual + brand, design systems, prototyping
Technical writing, marketing copy, PR, email outreach
Outbound, closing, pricing, customer success
Hiring, HR, finance, legal, admin
User research, market research, academic / scientific
Productivity, integration, meta-skills, personal automation
Career advice, skill-building, coaching, learning resources
Fresh off the press. Each entry links to a full report with finding-by-finding breakdown and the exact rule that triggered it.
Skills that earned an A or that taught us something about the threat landscape.
Six passes, four live layers. Static rules cover the easy stuff. Semantic, adversarial, and supply-chain passes are where the interesting findings come from.
Hard-coded regex and AST checks. Catches missing license, oversized files, secrets, malformed YAML, classic prompt-injection patterns.
An LLM reads SKILL.md the way a careful reviewer would. Catches ambiguous instructions, capability overreach, missing guardrails.
Fifteen attacks across five classes are run through a victim model. Findings only surface when at least two of three attempts in a class succeed.
Runs the skill once inside a sandbox with a mock LLM driver, records every file read or write, network fetch, and shell call, then audits the trace for claim-versus-behavior mismatches.
Sandbox follows every URL and import the skill references, fetches the actual content, and recursively audits it. Catches benign-looking pointers to high-risk payloads.
Parses every pip / npm dependency the skill declares, checks them against OSV.dev advisories, and flags typosquat candidates. Audit-only — no install. Surfaces SUP-001 typosquat / SUP-002 known CVE / SUP-003 unpinned dep findings.