blur-image

Original：🇺🇸 English
Translated

Use when the user wants to blur, redact, or anonymize parts of an image — screenshots with API keys, emails, PII, customer data, or any sensitive text. Also triggers on 'hide text in screenshot', 'redact image', 'blur sensitive', 'anonymize screenshot', 'prepare screenshot for sharing', or privacy-related image editing. Use this skill even if the user just says 'blur this' with an image file.

2installs
Sourcejamdesk/skills
Added on

NPX Install

npx skill4agent add jamdesk/skills blur-image

Tags

Translated version includes tags in frontmatter
image-blursensitive-data-protectionimagemagickscreenshot-editingprivacy-tool

SKILL.md Content

View Translation Comparison →

Blur Image

Detect and blur sensitive regions in screenshots using AI vision + ImageMagick.
Announce: "I'm using the blur-image skill to identify and blur sensitive regions."
Use when: Blur, redact, anonymize, hide text in screenshots, prepare images for sharing, remove PII from images.
Skip when: Crop, resize, format conversion, color adjustments, or non-privacy image editing.
Common mistake: Output goes to 
<name>-blurred.webp
by default, not overwriting the original. If the user expects in-place editing, they need 
--overwrite
.

Critical Rules

AlwaysNever
Check 
which magick
before running; if missing, provide install instructions and stop		Run blur without user confirmation
Show the generated command before executingOverwrite the original file without 
--overwrite
flag
Use sigma >= 15 for privacy-sensitive content (low sigma is reversible)Use sigma < 10 (can be reversed with deblurring algorithms)
Save output as WebP with 
-quality 85
Blur decorative/non-sensitive content without asking
Add 30-50px padding around detected regions, clamped to image edgesTrust raw AI coordinates without padding (20-50px error margins)
Read output image after blurring to verify correct regions were blurredSkip the verification step
Get image pixel dimensions with 
magick identify
before generating coordinates		Assume coordinates without checking actual image dimensions

Flags

FlagBehavior
(none)Full workflow: preflight, auto-detect, confirm, blur, verify
--guided
Skip auto-detection, ask user what to blur
--overwrite
Replace original file instead of creating 
-blurred
copy
--solid
Use 
-fill black -colorize 100
instead of Gaussian blur (more secure — can't be reversed)
--dry-run
Generate and show the command but don't execute it

Phase 1: Preflight

Check that ImageMagick is installed and the image is valid.
  1. Run 
    which magick
    . If missing, show install instructions and stop:
  2. Run 
    magick identify -format '%wx%h\n' image.png
    — this validates the file exists, format is supported (png, jpg, jpeg, webp, tiff), and returns pixel dimensions in one step. If the command outputs multiple lines, the image is animated (GIF, animated WebP, APNG) — stop and tell the user, because 
    -region
    blur only affects the first frame
  3. Report dimensions to confirm the coordinate space — Retina/HiDPI screenshots may be 2x logical resolution

Phase 2: Identify Regions

Two modes depending on flags and user input.

Mode A — Auto-detect (default)

  1. Read the image using multimodal vision
  2. Identify regions containing sensitive text:
    • Email addresses
    • API keys, tokens, secrets
    • URLs with credentials or internal hostnames
    • Personal names, phone numbers
    • IP addresses
    • File paths with usernames
    • Passwords, database connection strings
  3. For each region, determine the bounding box 
    (x, y, width, height)
    in pixels from top-left origin (ImageMagick's coordinate system)
  4. Add 30-50px padding to all sides of each detected region — AI spatial detection has 20-50px error margins. Clamp coordinates so regions don't extend past image edges (X >= 0, Y >= 0, X+W <= image width, Y+H <= image height)
  5. Present findings: 
    I found N regions with sensitive text:
1. [description] at roughly (X, Y) — WxH (with padding)
2. [description] at roughly (X, Y) — WxH (with padding)
  6. Ask user to confirm which regions to blur
  7. If no sensitive content found, report that and offer 
    --guided
    mode

Mode B — User-guided (
--guided
or when user specifies what to blur)

  1. User describes what to blur: "blur the email in the top right" or "blur everything below the header"
  2. Read the image, locate the described content
  3. Propose padded coordinates and ask user to confirm

Phase 3: Generate Command

Build the ImageMagick command with all confirmed regions.
Gaussian blur (default):
bash
magick input.png \
  -region WxH+X+Y -blur 0x20 \
  -region WxH+X+Y -blur 0x20 \
  -strip -quality 85 output-blurred.webp
Solid fill (
--solid
flag):
bash
magick input.png \
  -region WxH+X+Y -fill black -colorize 100 \
  -region WxH+X+Y -fill black -colorize 100 \
  -strip -quality 85 output-blurred.webp
  • Default output filename: 
    <original-name>-blurred.webp
    alongside the original (
    -strip
    removes EXIF metadata)
  • Present the full command and output path before running
  • Ask: "Run this command? The blurred image will be saved as [path]."

Phase 4: Execute and Verify

  1. Run the command
  2. Read the output image to verify blur was applied to the correct regions
  3. If any text is still visible or wrong area was blurred, offer to adjust coordinates (increase padding or shift position) and re-run
  4. Report success with the output file path

Example

User: blur the sensitive stuff in screenshot.png

Claude: I'm using the blur-image skill to identify and blur sensitive regions.

Preflight: magick found at /opt/homebrew/bin/magick (v7.1.2)
Image: screenshot.png (1920x1080 PNG)

I found 2 regions with sensitive text:
1. Email address "user@company.com" at roughly (430, 210) — 320x65px (with padding)
2. API key "sk_live_..." at roughly (80, 560) — 390x60px (with padding)

Blur both regions? [Y/n]

User: Y

Running:
  magick screenshot.png \
    -region 320x65+430+210 -blur 0x20 \
    -region 390x60+80+560 -blur 0x20 \
    -strip -quality 85 screenshot-blurred.webp

✓ Done. Saved to screenshot-blurred.webp
I've verified the output — both regions are blurred and surrounding content is intact.

Red Flags

Stop if you catch yourself:
  • Running blur without showing the command first
  • Using sigma < 10 (text may be recoverable with deblurring)
  • Overwriting the original without 
    --overwrite
    flag
  • Blurring the entire image instead of specific regions
  • Not reading the output image to verify the blur
  • Using coordinates without adding padding
  • Skipping 
    magick identify
    to confirm image dimensions
  • Not handling Retina/HiDPI resolution differences