FD agentskillshub
Use when the user wants to find, evaluate, audit, or install an open-source AI agent skill or MCP server — e.g. "find an MCP server for Postgres", "is this skill safe to install", "what should I use to scrape a website". Searches a quality-scored, security-graded catalog of ~20K skills locally (cached index, zero backend load) and returns each result's security grade, quality score, and install commands so you can check trust BEFORE installing.
g. "find an MCP server for Postgres", "is this skill safe to install", "what should I use to scrape a website". Searches a quality-scored, security-graded…
As a process D 45/100 · Unfinished process — weak spots: result and completion, inputs and preconditions, failures and branches
What is at stake
The skill contains fragments that, in the wrong hands, cost money or data. Below: what the installer risks and what the author should do.
The skill contains commands that delete files, rewrite disks or execute code fetched from the network. The agent may run them without asking if it believes the instructions require it.
Replace destructive commands with safe equivalents that ask for confirmation, scope them to one folder, and stop piping curl into a shell: pin a version and a checksum.
Below is the worst case for this category. The finding here is medium: the guard saw a sign, not a proof.
The instructions or scripts send data out: environment variables, keys, file contents, chat history. You may never notice, because the agent performs the upload as "part of the task".
If the upload is not needed for the task, remove it: catalogs flag such skills and delist them. If it is needed, name the destination explicitly, say what leaves the machine, and give the user a switch.
How to improve
- Address the high-severity findings: each costs 18 safety points. If one is a false positive, add the rule id to guard.allow in spec.yaml.
- Your own cases (evals/evals.json, 4–6 real requests with expected answers): the full check would then run those instead of a model-drafted suite.
- A spec.yaml with trigger phrases and assertions — a behaviour contract for CI; `skilltest init` writes a template.
Guard findings · 21
-
high Dangerous commands
cmd-encoded-execblog/securing-43k-ai-tools/index.html:253Executes a base64/encoded payload (documentation of a security skill)<tr><td>5</td><td>Dynamic Code Exec</td><td>Running obfuscated/dynamic code</td><td><code>base64 -d | bash</code></td></tr>
security skill
Medium and low: 20
-
medium Dangerous commands
cmd-pipe-to-shellblog/securing-117k-ai-skills/index.html:163Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<li><strong>Deeper, per-skill, on demand.</strong> Judging intent (is this <code>curl | sh</code> installing Homebrew, or exfiltrating a key?) and AST/taint-level analysis belong at the single-skill l
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-117k-ai-skills/index.html:171Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<li><strong>Semantic LLM review</strong> — an LLM judges intent on flagged skills (is this <code>curl | sh</code> installing Homebrew, or stealing a key?) to cut false positives. Built; being enabled.
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-117k-ai-skills/zh/index.html:153Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<li><strong>更深、单个、按需。</strong> 判断意图(这个 <code>curl | sh</code> 是在装 Homebrew,还是在外泄密钥?)和 AST/污点级分析,属于<strong>单个 skill</strong> 层面、按请求触发 —— 不对整个目录扫,否则太贵也太吵。这是按需 / 企业层,更严的检查住在这里。</li>
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-117k-ai-skills/zh/index.html:161Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<li><strong>LLM 语义复核</strong> —— 用 LLM 对被标记的 skill 判断意图(这个 <code>curl | sh</code> 是在装 Homebrew,还是在偷密钥?),降低误报。代码已具备,待启用。</li>
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-43k-ai-tools/index.html:214Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<p>It caught the obvious stuff. But it also flagged every legitimate tool that had <code>curl | bash</code> install instructions (which is... most of them). The false positive rate was terrible.</p>
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-43k-ai-tools/index.html:220Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)<p>The idea was simple: let the regex scanner do the fast first pass, then send flagged repos to an LLM for semantic analysis. The LLM could understand context — a <code>curl | bash</code> installing
security skill -
medium Dangerous commands
cmd-pipe-to-shellblog/securing-43k-ai-tools/index.html:225Downloads and executes remote code from an unrecognised host (pipe to shell) (documentation of a security skill)- curl|bash install scripts for well-known package managers
security skill -
medium Exfiltration
exfil-read-secret-filesblog/securing-43k-ai-tools/index.html:249Reads credential / secret files (documentation of a security skill)<tr><td>1</td><td>Data Exfiltration</td><td>Sending local data to external servers</td><td><code>curl -d $(cat ~/.ssh/id_rsa)</code></td></tr>
security skill -
medium Exfiltration
exfil-read-secret-filesblog/securing-43k-ai-tools/index.html:251Reads credential / secret files (documentation of a security skill)<tr><td>3</td><td>Sensitive Dir Access</td><td>Reading SSH keys, AWS credentials</td><td><code>cat ~/.aws/credentials</code></td></tr>
security skill -
low Obfuscation
obf-base64-blobblog/index.html:181Long base64-looking blob (quoted — discussed, not commanded)<form id="nl-form" onsubmit="return (function(e){e.preventDefault();var em=document.getElementById('nl-email').value;if(!em)return false;var btn=document.getElementById('nl-btn');btn.textContent='Subsquoted -
low Dangerous commands
cmd-privilegeblog/securing-43k-ai-tools/index.html:254Privilege escalation / world-writable permissions (documentation of a security skill)<tr><td>6</td><td>Privilege Escalation</td><td>Gaining root access</td><td><code>chmod 777</code>, <code>chown root</code></td></tr>
security skill -
low Dangerous commands
cmd-shell-rcblog/securing-43k-ai-tools/index.html:255Writes to a shell startup file (documentation of a security skill)<tr><td>7</td><td>Persistence</td><td>Surviving reboots via cron/bashrc</td><td><code>crontab</code>, <code>>> ~/.bashrc</code></td></tr>
security skill -
low Obfuscation
obf-base64-blobblog/zh/index.html:162Long base64-looking blob (quoted — discussed, not commanded)<form id="nl-form" onsubmit="return (function(e){e.preventDefault();var em=document.getElementById('nl-email').value;if(!em)return false;var btn=document.getElementById('nl-btn');btn.textContent='订阅中.quoted
A further 7 matches are quotations in this security skill's documentation and are not counted as findings.
Files scanned: 22. Evidence is masked. Grey chips explain why severity was lowered.
Against the Agent Skills spec
✓ No remarks against the Agent Skills spec
Process rating: all ten parameters 45/100
- 0Result and completion. Does not say what the result is
- 0Inputs and preconditions. Does not say what the process needs to start
- 0Failures and branches. Linear process with no failure handling
- 0Progress reporting. Says nothing while it works
- 30Running it twice. 1 mutating operations with no state check
- 40Consistency. Frontmatter name (agentskillshub) differs from the folder (public)
- 60Tools and files. Uses tools (bash, node) that frontmatter does not declare
- 70When it triggers. States when to use, but not when not to
- 100Steps. 12 steps
- 100Execution cost. Instruction body is 884 tokens
Everything here is measured from the skill text rather than judged by a model, so the numbers are checkable. A parameter weighs more when it is a more common reason for the process to stall.
Quality signals
- +4Description does not say when NOT to use the skill (false activations)
- +3Output format is not stated: the model decides each time
- +1No license
- +2Single-language instructions
- +5Description quotes 3 example trigger phrases
- +3Description length 448: enough signal without eating the budget
- +4Structure: 6 headings
- +3Step-by-step instructions: 12 items
- +4Has examples (1 code blocks)
Quality base 70; lint remarks subtract, signals add up to 100. Result: 89.