Methodology
560 Open-Source Agent Skills: Deep Deconstruction & Control Level Auditing
1Core Philosophy & Definition
Unified Definition:
Skills run on user hosts and are executed by LLMs and agent runtimes. Our control levels indicate the blast radius and reach of each skill, suggesting appropriate authorization, isolation, and audit controls. A higher level implies stronger governance requirements, not poor quality or forbidden use. Different agent hosts apply varying constraints; never assume a host provides automatic safeguards. Third-party skills belong to their original creators; usage risks are borne by the user.
Skills run on user hosts and are executed by LLMs and agent runtimes. Our control levels indicate the blast radius and reach of each skill, suggesting appropriate authorization, isolation, and audit controls. A higher level implies stronger governance requirements, not poor quality or forbidden use. Different agent hosts apply varying constraints; never assume a host provides automatic safeguards. Third-party skills belong to their original creators; usage risks are borne by the user.
2Control Level Governance Matrix (L1–L5)
| Level | Characteristics | Agent Recommendation | Admin Governance |
|---|---|---|---|
| L1 Basic | Pure prompt instructions, no scripts, no external access | Can be auto-loaded | Register source and version |
| L2 Local | Read/write local workspace files, constrained or built-in deterministic scripts | Restrict working directory | Retain operation logs |
| L3 Networked | Network access, external API service calls, requires API keys | Declare allowed domains and key scope | Least privilege keys, retain request logs |
| L4 High-Privilege | Arbitrary command execution, dependency installation, remote resource mutation, data exfiltration | Graded/per-use authorization, sandbox recommended | Approval workflow, rollback support, full audit |
| L5 Restricted | Session cookies, reverse/private APIs, anti-bot bypass, platform ToS/compliance risk | Isolated environment, per-use explicit authorization | Strict authorization, full audit, compliance signoff |
3Capability Tags & Thresholds
Reverse API Min L5
Key capability identifier triggering L5 governance threshold.
Terms / Compliance Min L5
Key capability identifier triggering L5 governance threshold.
Credential / Session Min L5
Key capability identifier triggering L5 governance threshold.
Execute Command Min L4
Key capability identifier triggering L4 governance threshold.
Install Deps Min L4
Key capability identifier triggering L4 governance threshold.
Data Exfiltration Min L4
Key capability identifier triggering L4 governance threshold.
External Write Min L4
Key capability identifier triggering L4 governance threshold.
Network Access Min L3
Key capability identifier triggering L3 governance threshold.
Write File Min L2
Key capability identifier triggering L2 governance threshold.
4Disclaimer & Terms
Skills run on user hosts and are executed by LLMs and agent runtimes. Our control levels indicate the blast radius and reach of each skill, suggesting appropriate authorization, isolation, and audit controls. A higher level implies stronger governance requirements, not poor quality or forbidden use. Different agent hosts apply varying constraints; never assume a host provides automatic safeguards. Third-party skills belong to their original creators; usage risks are borne by the user.