Skill Watch · rules 2026-10-19.3
What skills tell agents to do
Public skills that matched a risky pattern, newest first. Each links to the exact lines. A match is a reason to read, not a verdict on the author.
- 9,931 skills scanned
- 77 with a high-risk pattern
- 194 with a medium-risk pattern
Counts include security courses and scanner test files, which quote attacks on purpose, and results we hold back while their maintainers hear from us. The launch report separates them: 30 of 9,014 ordinary skills matched a high-risk pattern.
"Flagged" here means the scanner matched a high-risk text pattern; it is not a Watchlist flag, which needs three verified people.
Most common patterns
Newest
| Verdict | Skill | Top pattern | Seen |
|---|---|---|---|
| High-risk pattern | libafl trailofbits/skills | A download piped straight into a shell | 2026-10-19 |
| High-risk pattern | developing-genkit-go google/skills | A download piped straight into a shell | 2026-10-19 |
| High-risk pattern | developing-genkit-dart google/skills | A download piped straight into a shell | 2026-10-19 |
| High-risk pattern | verify anthropics/claude-agent-sdk-python | A download piped straight into a shell | 2026-10-19 |
| High-risk pattern | vss-deploy-profile NVIDIA/skills | A download piped straight into a shell | 2026-10-19 |
| High-risk pattern | tao-run-on-brev NVIDIA/skills | A download piped straight into a shell | 2026-10-19 |
| Medium-risk pattern | yara-rule-authoring trailofbits/skills | Reads a secret from the system keychain | 2026-10-19 |
| Medium-risk pattern | interpreting-culture-index trailofbits/skills | A download piped straight into a shell | 2026-10-19 |
| Medium-risk pattern | agentic-actions-auditor trailofbits/skills | Turns off the agent's permission prompts | 2026-10-19 |
| Medium-risk pattern | gke-compute-classes google/skills | An instruction to keep something from the user | 2026-10-19 |
| Medium-risk pattern | vss-generate-video-report NVIDIA/skills | Collects environment variables or tokens | 2026-10-19 |
| Medium-risk pattern | tao-setup NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-on-virtualenv NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-on-slurm NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-on-kubernetes NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-on-docker NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-inference-service NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-run-deft-aoi NVIDIA/skills | Changes the agent's standing files or memory | 2026-10-19 |
| Medium-risk pattern | tao-finetune-cosmos-embed NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | tao-data-io NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | rtvi-vlm-customize-model NVIDIA/skills | Collects environment variables or tokens | 2026-10-19 |
| Medium-risk pattern | physical-ai-defect-image-generation NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | paidf-orchestration-setup NVIDIA/skills | Collects environment variables or tokens | 2026-10-19 |
| Medium-risk pattern | paidf-cosmos-predict NVIDIA/skills | Reads a .env file | 2026-10-19 |
| Medium-risk pattern | nemotron-customize NVIDIA/skills | Collects environment variables or tokens | 2026-10-19 |
| Medium-risk pattern | nemo-mbridge-perf-cuda-graphs NVIDIA/skills | Instructions inside an HTML comment | 2026-10-19 |
| Medium-risk pattern | hsb-test NVIDIA/skills | Turns off the agent's permission prompts | 2026-10-19 |
| Medium-risk pattern | hsb-setup NVIDIA/skills | Turns off the agent's permission prompts | 2026-10-19 |
| Medium-risk pattern | hsb-app NVIDIA/skills | Turns off the agent's permission prompts | 2026-10-19 |
| Medium-risk pattern | amc-run-rtsp-calibration NVIDIA/skills | Collects environment variables or tokens | 2026-10-19 |
Scan a skill · The rules · Follow (Atom feed) · JSON
Check your own skills
Or paste this into your agent once, and it checks every skill before installing it:
Read https://theagentforge.io/skill.md and use it: before you install any skill, plugin or MCP server, check it with the Forge.Free, no account. Read the skill file first.