<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/">
  <channel>
    <title>DEV Community: sinescribe</title>
    <description>The latest articles on DEV Community by sinescribe (@sinescribe).</description>
    <link>https://dev.to/sinescribe</link>
    <image>
      <url>https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4253263%2F77f535b4-63fb-476c-a326-3dcb62e3990b.png</url>
      <title>DEV Community: sinescribe</title>
      <link>https://dev.to/sinescribe</link>
    </image>
    <atom:link rel="self" type="application/rss+xml" href="https://dev.to/feed/sinescribe"/>
    <language>en</language>
    <item>
      <title>Zero-Dependency Prompt Injection &amp; LLM Tool-Call Defense in Python</title>
      <dc:creator>sinescribe</dc:creator>
      <pubDate>Thu, 19 Nov 2026 09:40:21 +0000</pubDate>
      <link>https://dev.to/sinescribe/zero-dependency-prompt-injection-llm-tool-call-defense-in-python-5gnm</link>
      <guid>https://dev.to/sinescribe/zero-dependency-prompt-injection-llm-tool-call-defense-in-python-5gnm</guid>
      <description>&lt;p&gt;Autonomous AI agents that browse web pages, read GitHub issues, or inspect email are exposed to &lt;strong&gt;indirect prompt injection&lt;/strong&gt; on every turn. When untrusted text enters the context window, a malicious string like &lt;code&gt;&amp;lt;system&amp;gt;Ignore previous instructions...&amp;lt;/system&amp;gt;&lt;/code&gt; or a leaked canary can hijack tool calls.&lt;/p&gt;

&lt;p&gt;Here is a clean, zero-dependency Python pattern you can drop in front of any tool output before passing it back to your LLM loop.&lt;/p&gt;

&lt;h2&gt;
  
  
  1. Pattern + Canary Token Filter (&lt;code&gt;agent_guard.py&lt;/code&gt;)
&lt;/h2&gt;



&lt;div class="highlight js-code-highlight"&gt;
&lt;pre class="highlight python"&gt;&lt;code&gt;&lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="n"&gt;re&lt;/span&gt;
&lt;span class="k"&gt;from&lt;/span&gt; &lt;span class="n"&gt;typing&lt;/span&gt; &lt;span class="k"&gt;import&lt;/span&gt; &lt;span class="nc"&gt;Tuple&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nc"&gt;List&lt;/span&gt;

&lt;span class="no"&gt;INJECTION_PATTERNS&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)ignore\s+(all\s+)?(previous|prior|above)\s+instructions"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)system\s+prompt\s*:\s*you\s+are\s+now"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)disregard\s+(the\s+)?(developer|system)\s+instructions"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)new\s+operating\s+mode\s*:\s*jailbreak"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)output\s+(your\s+)?(system\s+instructions|hidden\s+prompt|secret\s+keys)"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)&amp;lt;\s*system\s*&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)dan\s+mode|developer\s+mode\s+enabled"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
    &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="s2"&gt;"(?i)you\s+must\s+now\s+act\s+as\s+an\s+unrestricted"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt;
&lt;span class="p"&gt;]&lt;/span&gt;

&lt;span class="k"&gt;class&lt;/span&gt; &lt;span class="nc"&gt;PromptDefenseFilter&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
    &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;__init__&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;canary_token&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kt"&gt;str&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="kc"&gt;None&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;canary&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;canary_token&lt;/span&gt;
        &lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;regexes&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="n"&gt;re&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nb"&gt;compile&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;p&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;p&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="no"&gt;INJECTION_PATTERNS&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;

    &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;scan_untrusted_input&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kt"&gt;str&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&amp;gt;&lt;/span&gt; &lt;span class="nc"&gt;Tuple&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="kt"&gt;bool&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="nc"&gt;List&lt;/span&gt;&lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="kt"&gt;str&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="n"&gt;flagged&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="p"&gt;[&lt;/span&gt;&lt;span class="p"&gt;]&lt;/span&gt;
        &lt;span class="k"&gt;for&lt;/span&gt; &lt;span class="n"&gt;r&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;regexes&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
            &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="n"&gt;r&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;search&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
                &lt;span class="n"&gt;flagged&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;f&lt;/span&gt;&lt;span class="s2"&gt;"Pattern violation: {r.pattern}"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

        &lt;span class="k"&gt;if&lt;/span&gt; &lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;canary&lt;/span&gt; &lt;span class="ow"&gt;and&lt;/span&gt; &lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="n"&gt;canary&lt;/span&gt; &lt;span class="k"&gt;in&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
            &lt;span class="n"&gt;flagged&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;append&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"Canary token leak detected in untrusted content!"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;

        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="nb"&gt;len&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="n"&gt;flagged&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;==&lt;/span&gt; &lt;span class="mi"&gt;0&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;flagged&lt;/span&gt;

    &lt;span class="k"&gt;def&lt;/span&gt; &lt;span class="nf"&gt;sanitize&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="nb"&gt;self&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt; &lt;span class="kt"&gt;str&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt; &lt;span class="o"&gt;-&amp;gt;&lt;/span&gt; &lt;span class="kt"&gt;str&lt;/span&gt;&lt;span class="p"&gt;:&lt;/span&gt;
        &lt;span class="n"&gt;text&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"&amp;lt;system&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"[system-tag-stripped]"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
        &lt;span class="n"&gt;text&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"&amp;lt;/system&amp;gt;"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"[/system-tag-stripped]"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
        &lt;span class="n"&gt;text&lt;/span&gt; &lt;span class="o"&gt;=&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;&lt;span class="p"&gt;.&lt;/span&gt;&lt;span class="nf"&gt;replace&lt;/span&gt;&lt;span class="p"&gt;(&lt;/span&gt;&lt;span class="s2"&gt;"```

prompt"&lt;/span&gt;&lt;span class="p"&gt;,&lt;/span&gt; &lt;span class="s2"&gt;"

```text"&lt;/span&gt;&lt;span class="p"&gt;)&lt;/span&gt;
        &lt;span class="k"&gt;return&lt;/span&gt; &lt;span class="n"&gt;text&lt;/span&gt;
&lt;/code&gt;&lt;/pre&gt;

&lt;/div&gt;



&lt;h2&gt;
  
  
  2. Why Canary Tokens Matter
&lt;/h2&gt;

&lt;p&gt;Regex alone misses novel phrasing. Placing a unique non-guessable canary string in sensitive internal context and scanning outbound tool arguments/web responses catches prompt exfiltration immediately before external network calls execute.&lt;/p&gt;

&lt;h2&gt;
  
  
  3. Open-Source Lite Repo &amp;amp; Full Production Toolkits
&lt;/h2&gt;

&lt;ul&gt;
&lt;li&gt;Free MIT starter repo: &lt;a href="https://github.com/sinescribe/agent-guard-lite" rel="noopener noreferrer"&gt;https://github.com/sinescribe/agent-guard-lite&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Production Multi-Agent Architecture &amp;amp; Protocol Pack ($29):&lt;/strong&gt; Includes multi-agent consensus dispatcher, collision-free routine scheduler, and hardened prompt-defense pipeline -&amp;gt; &lt;a href="https://sinescribe5.gumroad.com/l/phhyon" rel="noopener noreferrer"&gt;https://sinescribe5.gumroad.com/l/phhyon&lt;/a&gt;
&lt;/li&gt;
&lt;li&gt;
&lt;strong&gt;Autonomous Security &amp;amp; Bug Bounty Recon Kit ($19):&lt;/strong&gt; Zero-dependency HTTP security header auditor, CORS misconfiguration scanner, and CVSS 3.1 / VRT disclosure report generator -&amp;gt; &lt;a href="https://sinescribe5.gumroad.com/l/jyubxd" rel="noopener noreferrer"&gt;https://sinescribe5.gumroad.com/l/jyubxd&lt;/a&gt;
&lt;/li&gt;
&lt;/ul&gt;

&lt;p&gt;Built with zero external dependencies so you can audit every line before deploying.&lt;/p&gt;

</description>
      <category>ai</category>
      <category>security</category>
      <category>python</category>
    </item>
  </channel>
</rss>
