Scan a webpage before giving it to an AI assistant to catch hidden prompt injection or instruction smuggling.
Review HTML/HTM, SVG, MD/Markdown, marketing pages, or public websites for hidden content, comments, and webpage metadata that may target AI agents.
Use strict mode when redirect boundaries matter, and carry any partial-coverage notice into the decision.
Document review
Before a file is uploaded or shared
Check a PDF from email, cloud storage, or messaging apps for invisible or misleading text before sharing it with an LLM.
Inspect DOCX comments, revisions, vanished text, and metadata before an assistant processes the document.
Review EML alternative bodies and supported textual attachments, ICS descriptions, or XLSX hidden sheets and suspicious formulas.
Inspect PNG, JPG/JPEG, WebP, or GIF images for supported visual signals and QR payloads; animated images are limited to the first frame.
Use the app as a local privacy-preserving scanner when content should not be uploaded to a remote service.
Reports and evidence
When a scan result needs to be documented clearly
Generate a clear risk report for suspicious files or links so a user can decide whether to trust, share, or upload them.
Export findings as JSON or PDF for internal review, compliance evidence, or incident tracking.
Help security or trust teams document suspicious content with structured evidence instead of ad hoc screenshots.
Team and pipeline review
When content feeds a larger AI workflow
Send a link, file, or selected text through the Share Sheet for a compact review before it enters the next tool.
Triage third-party documents before they are added to a knowledge base, RAG corpus, or agent memory system.
Screen supplier, client, or contractor documents before they are processed by internal AI tools.
Scan resumes, cover letters, and candidate portfolio URLs before AI-assisted hiring review or ATS automation runs.
Evaluation and testing
When the goal is to harden systems before production
Audit training or benchmark datasets for prompt-injection-like artifacts before using them in evaluation pipelines.
Support QA and red-team workflows by testing whether known prompt-injection samples are detected correctly.
Exercise Base64, hexadecimal, URL-percent, ROT13, and split-artifact cases against the layered on-device detector.
Evaluate selected Spanish, Chinese, and Arabic patterns without implying coverage for every language.
Premium local automation
When the scan should plug into a repeatable local workflow
Use premium scan_text, scan_url, scan_file, or scan_repo while the local app runtime is active.
Retrieve redacted reports or apply runtime guards before an agent proceeds with a risky plan or action.
MCP workflows
Use MCP when the scan should stay inside an agent workflow.
Automation runs in an active local session. It is built for agent workflows that need more than a single manual scan:
text, URL, file, or repository inspection, report retrieval, and plan or action gating.
Local scan-and-report loop
Open a session, choose scan_text, scan_url, scan_file, or scan_repo, retrieve or export the report, then close the session.
Bounded repository intake
Use scan_repo before an agent reads or installs from a local repository. It is an input-hardening signal, not a complete repository-security verdict.
Guard plans and actions
Use scope_tools, guard_plan, and gate_action when the scanner needs to participate in runtime policy checks, not just detection.
Choose the scan path that matches how the content will be used.
The app supports the supported formats behind these scenarios: Share Sheet intake, supported files and images,
user-triggered HTTPS scans, reviewed reports, redacted exports, and supported local automation.
Veridicus Scan