Guides for building skills that hold up.
Guides and editorial write-ups from the library on the decisions, trade-offs, and engineering work behind dependable agent skills.
Spectrum of skill sophistication
A spectrum of speed, variability, rigor, and operational maturity.
Read the guideMore guides from the library
Practical reading for deciding whether to use, install, review, or improve an agent skill.
Langfuse CLI beta: security review
Review the Langfuse CLI beta’s file, credential, and update controls, with scoped findings, scanner limitations, and evidence for a constrained pilot.
Resource library rating methodology
Understand the five usefulness criteria, inspect the evidence behind every resource's score, and learn how ratings are reassessed.
Skill security over time
Assess MCP-loaded skills, match audits to exact versions, and keep a practical record of updates, permissions, testing, and rollback.
AI evals: a quick guide
Start with human trace review, choose practical evaluation checks, and inspect scoped security results for all nine evals-skills packages.
Literature review of Skills Efficacy
A research synthesis on personalized Skills, task-matched guidance, retrieval, plugin packaging, cost, and security—with the full supporting notes.
Enterprise training on skills for AI agents: presenter readiness questions
A presenter-facing question bank for preparing a clear, credible 20-minute enterprise introduction to agent skills.
Skills in the agentic performance system
A seven-lever framework for seeing Skills alongside models, data, tools, harnesses, software, evaluation, and trust.
Controlled variation: the capability and control problem of AI agents
Why agent flexibility creates both novel problem-solving value and reliability risk—and how Skills, scripts, guardrails, and evals control the tradeoff.
Agentic performance source library
An annotated, source-labeled reading map spanning AAIF, major providers, research papers, and O'Reilly Radar.
Writing without the AI sheen
A practical authenticity-first editing workflow, with static safety evidence for four community writing skills.
From images to editable design
Compare Ming layers, Ling UI, and editable PowerPoint workflows, with scoped security evidence and a practical path from images to editable design.
When not to use a skill
A decision guide for choosing primary documentation or a governed internal workflow instead of a third-party agent skill.
So you found a skill: pre-install confidence checklist
A source-backed checklist for reviewing a skill's provenance, behavior, dependencies, permissions, reliability, and update path before installation.
NVIDIA SkillSpector enterprise training guide
A practical guide to screening an agent skill before installation and turning scanner output into a defensible human decision.
Reviewing UI/UX Pro Max with NVIDIA SkillSpector
A case study in reviewing a skill bundle with static analysis, OWASP prompt-injection triage, and a scoped semantic second opinion before installation.
Obra Superpowers: 14 skills and a SkillSpector review
A commit-pinned map of Obra's 14-skill workflow, with per-skill static findings, JSON evidence, and the archived research thread that started the review.
Matt Pocock’s Skills for Real Engineers: all 37 reviewed
A commit-pinned, skill-by-skill review of Matt Pocock’s modular engineering collection, with SkillSpector evidence, community feedback, and a direct comparison to Obra Superpowers.
What an agent skill looks like: SkillSpector Review
An annotated walkthrough of a real Codex skill that turns a security policy into a discoverable, reusable workflow.
SkillSpector Review v2: a layered pre-install gate
A walkthrough of the current global skill: registry evidence, exact-artifact static analysis, OWASP triage, and authorization-aware semantic review.
skills.sh Security Check: collect three audit reports
Download a Codex skill that finds the current Gen Agent Trust Hub, Socket, and Snyk evidence for a third-party skill without installing it first.
Microsoft SkillOpt: third-party developer snapshot
An independent developer review of SkillOpt's maturity, operating costs, security boundaries, and fit for governed self-service skills.
Headroom: effectiveness and security evidence
Compare Headroom's token claims with third-party coding-agent tests, a scoped SkillSpector review, and the separate skills.sh community-skill audits.
JetBrains tests Caveman: real token savings
A reading guide to JetBrains' paired benchmark of terse agent narration, where the advertised 65% saving became 8.5% on real coding-agent work.
JetBrains tests rtk: compression versus the bill
A reading guide to JetBrains' rtk trial and its warning that a tool's token counter can rise while the measured agent bill does too.
JetBrains tests Ponytail: measured minimal code
A reading guide to JetBrains' Ponytail benchmark, where minimal-code guidance cut emitted code and cost most on over-built tasks.
Anthropic’s Complete Guide For Building Skills For Claude [VERBATIM]
Anthropic’s original 33-page guide, presented through the publisher-hosted PDF with clear source attribution and no editorial rewriting.