alpha · v0.3.0

Find what's hiding in plain sight.

Invisible Unicode, homoglyphs, whitespace channels, watermark signals — StegGuard finds them, decodes them, and writes a clean copy. Runs on the Python standard library. Offline by default.

pip install git+https://github.com/stegguard/stegguard

Read the docs

leaked-prompt.md

$ stegguard detect leaked-prompt.md

One shared workflow, instead of five one-off tools.

The original file stays untouched unless you explicitly authorize an in-place replacement.

Detect. Point it at a file, a folder, or a whole repo.

Runs on the standard library. No network calls, no upload. Exits 1 when something is flagged.

$ stegguard detect

$ stegguard detect report.md
 
📄 report.md
  ⚠  ZERO-WIDTH CHARACTERS DETECTED — Severity: MEDIUM
  …
    Line 2  Col 7  Char #15  U+200B  Zero Width Space
    Line 2  Col 8  Char #16  U+200C  Zero Width Non-Joiner
  …
  ⚠ 1/1 file(s) flagged
exit 1

Decode. Extract the payload without touching the source.

Prints a JSON report. The input file is never modified.

$ stegguard decode

$ stegguard decode report.md
…
  "status": "DECODED",
  "zero_width": {
    "decoded": "hi",
    "confidence": 0.9,
    "occurrences": 16
  },
…

Sanitize. Write a clean copy and get a change report.

Writes report.sanitized.md by default. Overwriting the original needs --in-place and --confirm, together.

$ stegguard sanitize

$ stegguard sanitize report.md
…
  "sanitized_path": "report.sanitized.md",
  "sha256_before": "6c39d823…",
  "sha256_after": "a5d49ca2…",
  "changes": {
    "zero_width_removed": 16,
    "bidi_removed": 1,
    …
  },
…
  "original_preserved": true,
  "sanitization_status": "SANITIZED"

Hidden content is rarely one neat payload in one neat format.

Text & source	Zero-width Unicode, bidirectional overrides, Cyrillic/Greek homoglyphs, trailing-space and tab channels, mixed line endings.	any text or source file
Images, audio & media	Pixel-level LSB anomalies (chi-square, RS, entropy, sample-pair), PNG private chunks, JPEG EXIF/XMP/IPTC, PCM audio statistics.	images, audio, structured media
Documents & containers	PDF embedded files, DOCX/PPTX/XLSX/ZIP members, nested archives, SVG and HTML hidden content.	documents, archives
Watermarks & provenance	Metadata, structural, and text-pattern evidence; device fingerprints; C2PA manifests validated through c2patool.	images, audio, video

Built for the moment untrusted content meets your pipeline.

CI gate: Block pull requests that smuggle BIDI overrides or homoglyphs into source files before they merge.
LLM input sanitization: Catch zero-width prompt-injection payloads hiding in pasted text, scraped pages, or model output.
Pre-publish checks: Scan DOCX, PPTX, and PDF files for hidden fields and layout tricks before they go out the door.
Provenance audits: Validate C2PA Content Credentials and flag robust-watermark signals before trusting an image or audio file.

Everything on this page ships in the Apache-2.0 open-source package.

Read the full docs →