Semantic Phishing Honeybot

SkillSecurity

Offline experimental text-turn helper for phone scam-review demonstrations. Extracts candidate IoCs and suggests bait responses; it does not place, intercept, or answer calls.

Available today. Use it from your connected AI after setup.

Add ahel to your AI once: Claude, ChatGPT, Cursor, Claude Code or Codex. Then ask it to use this.

Then ask your AI: use the Semantic Phishing Honeybot skill

What this skill tells your AI

The instructions your AI receives, as published by calle-ai/awesome-phone-call-agents in skills/call-semantic-phishing-honeybot/SKILL.md and read by ahel’s review.

The supplied helper uses regex over synthetic transcript text and returns candidate URLs, wallet strings, phone-like strings, and predefined response suggestions. It does not determine that a person is a scammer, operate a call, browse extracted URLs, or send threat reports. Any future engagement must be operator-authorized, bounded, stoppable, and separate from this offline demonstration.

Scientific Foundation

Paper / ConceptRelevance
Active Defense & HoneybotsEmail scam-baiting research motivates this design; it does not establish phone-call duration or effectiveness.
Social Engineering TaxonomyUses established psychological countermeasures (acting confused, feigning compliance) to trigger the scammer into repeating technical instructions (IoCs).
Vishing Kill-Chain AnalysisTargets the "Action/Exploitation" phase of the voice phishing kill chain to capture the final payload (e.g. the Bitcoin wallet).

How it works

  1. Supply a synthetic transcript turn and fictional caller identifier; no call is answered or intercepted.
  2. The helper applies regex to the text to extract candidate indicators.
  3. It dynamically generates a bait_prompt tailored to the scammer's current objective:
    • If the scammer wants Bitcoin, the bot feigns ignorance about crypto to drag out the call.
    • If the scammer provides a URL, the bot pretends its antivirus blocked it to gather alternative domains.
  4. Candidates are returned in HoneybotResponse; they are neither verified nor automatically logged/shared.

Expected Outcomes & Metrics

These are unvalidated design targets for a future integration, not measurements or effects of the local helper.

MetricTargetNotes
IoC Extraction Rate> 75%Successfully pulling a wallet or URL from a known scam script.
Call Duration (Stalling)> 5 minsTime wasted per scammer.

Limitations & Known Constraints

  • Pattern Matching Limits: The current IoC extraction relies on basic regex (e.g. base58 for Bitcoin). It may miss newer or obfuscated wallet formats.
  • Integration Boundary: No LLM or calling loop is included. A future host must impose an operator-controlled stop and duration limit; returned continue_baiting is a suggestion, not authority to continue a call.

Signals

GitHub stars
104
Forks
527
Last commit
Sep 2026
Advanced
Item type
skill
Key
call-semantic-phishing-honeybot
Source
github.com/calle-ai/awesome-phone-call-agents