Share-Safe Renders
Find what a render leaks before you share it: API keys, Discord and Slack webhooks, user names in paths, e-mails, LAN IPs, GPS and hidden MP4 data. The report never prints the leaked text.
Free
Works with the AI Agents you already use
Share-Safe Renders
Example session with this skill installed
I'm going to post the six renders in renders/ on a forum, one at a time. Can you check them first? (The folder holds the six demo files from scripts/make_demo.py, and the page DEMO_EXPECTED.md that comes with them.)
- Read your context and instructions
- Compiled the share-safe renders
- Generated the image
Demo files: the six synthetic files that scripts/make_demo.py writes into renders/ (invented keys, an invented webhook, a made-up user name, address and coordinates). The numbers, places and hashes below are the scanner's real, unedited output (scan.py 1.2.0); the layout is the one SKILL.md prescribes. Nothing was sent or published, and no file was changed.
Verdict. Do not share these files as they are: 12 finding(s) in 5 of 6 file(s) (7 high, 4 medium, 1 low).
Findings (same value in two places is one secret, so those rows are merged)
| Severity | Class | Meaning | File | Where |
|---|---|---|---|---|
| high | K2 + K1 | a value under a secret-looking field name (api_key, token, password, secret, authorization ...); a string shaped like a vendor API key or token (OpenAI, Anthropic, GitHub, AWS, Google, Slack, Stripe live, Hugging Face) | renders/01_png_api_key_and_username.png | png:tEXt, tEXt[prompt]:node:11/inputs/api_key, node 11 |
| high | K3 + H3 | a Discord or Slack webhook URL | renders/02_png_ztxt_webhook.png | png:zTXt, zTXt[workflow]:nodes[1]/widgets_values[0], node 7 |
| high | K2 + K1 + H1 | a value under a secret-looking field name (api_key, token, password, secret, authorization ...); a string shaped like a vendor API key or token (OpenAI, Anthropic, GitHub, AWS, Google, Slack, Stripe live, Hugging Face) | renders/03_png_bypassed_node_secret.png | png:tEXt, tEXt[workflow]:nodes[2]/widgets_values[0]/api_key, node 12, bypassed |
| high | K1 | a string shaped like a vendor API key or token (OpenAI, Anthropic, GitHub, AWS, Google, Slack, Stripe live, Hugging Face) | renders/05_mp4_prompt_tag_and_free_box.mp4 | mp4:box:free, free:raw |
| high | G1 | GPS coordinates (EXIF, XMP, an MP4/MOV location tag or the MP4 loci box) | renders/06_jpeg_exif_gps.jpg | jpeg:exif, exif:GPS |
| medium | P1 | a home-folder path that reveals an account name (Windows, macOS and Linux forms) | renders/01_png_api_key_and_username.png | png:tEXt, tEXt[prompt]:node:9/inputs/image, node 9 |
| medium | P2 | an e-mail address (documentation addresses such as example.com are ignored) | renders/05_mp4_prompt_tag_and_free_box.mp4 | mp4:tag, moov/udta/meta/ilst/prompt:node:20/inputs/~b0a1bd37, node 20 |
| medium | M1 | hidden MP4/MOV data: a tag holding workflow or prompt JSON, or data in a free/uuid/unknown top-level box or after the last box | renders/05_mp4_prompt_tag_and_free_box.mp4 | mp4:tag, moov/udta/meta/ilst/prompt |
| medium | M1 | hidden MP4/MOV data: a tag holding workflow or prompt JSON, or data in a free/uuid/unknown top-level box or after the last box | renders/05_mp4_prompt_tag_and_free_box.mp4 | mp4:box:free, free |
| low | P3 | a private-network address (10.x, 172.16-31.x, 192.168.x) or a .local/.lan/.internal host name | renders/05_mp4_prompt_tag_and_free_box.mp4 | mp4:tag, moov/udta/meta/ilst/prompt:node:20/inputs/server, node 20 |
4 PNG file(s) carry an embedded ComfyUI workflow. The prompt text, model names, seeds and node layout travel with the file although they are not findings.
The scanner did not examine 1 file(s) in the folder because they are not renders: renders/DEMO_EXPECTED.md.
How to fix
- Export a copy of each flagged file without its metadata, into a new folder (the originals stay untouched). Every file name here is plain (letters, digits,
_and.), so the commands can be built from them; a file whose name is not plain would get no command, only a request to rename it ("Names in commands" inSKILL.md):
mkdir reexported
ffmpeg -loglevel error -n -i "renders/01_png_api_key_and_username.png" -map_metadata -1 "reexported/01_png_api_key_and_username.png"
ffmpeg -loglevel error -n -i "renders/02_png_ztxt_webhook.png" -map_metadata -1 "reexported/02_png_ztxt_webhook.png"
ffmpeg -loglevel error -n -i "renders/03_png_bypassed_node_secret.png" -map_metadata -1 "reexported/03_png_bypassed_node_secret.png"
ffmpeg -loglevel error -n -i "renders/05_mp4_prompt_tag_and_free_box.mp4" -map_metadata -1 -map_chapters -1 -c copy "reexported/05_mp4_prompt_tag_and_free_box.mp4"
ffmpeg -loglevel error -n -i "renders/06_jpeg_exif_gps.jpg" -map_metadata -1 -q:v 2 "reexported/06_jpeg_exif_gps.jpg"
- Run those and tell me when the 5 copies exist. I will then scan them (
python "<SKILL_DIR>/scripts/scan.py" "reexported" --json "share-safe-report-2.json") and check that the number of files scanned is 5. - The files with a key or webhook finding (K1, K2, K3): if they were ever posted or sent before, treat that key or webhook as exposed and revoke it at the provider. Removing metadata does not take it back.
What this scan cannot see
- Pixel-hidden data and what the picture shows
- Data inside IDAT, JPEG entropy data and the first mdat, ICC profile contents and UTF-16 text in raw regions
- Contents of C2PA manifests
- Secrets split, re-encoded or glued to a letter, placeholder values, unrecognised secret field names, vendor key formats outside its eight and webhooks other than Discord and Slack
- Formats it cannot open (WebP, GIF, WebM, HEIC) and addresses other than 10.x, 172.16-31.x and 192.168.x (IPv6, loopback, link-local, CGNAT, public)
01_png_api_key_and_username.png
PNG · 1536×1024
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
Find out what a render leaks before you share it. Share-Safe Renders gives your agent one scanner for PNG, JPEG and MP4/MOV files and ComfyUI workflow JSON. It reports what people post by accident: API keys and secret-named fields, Discord and Slack webhook URLs, user names inside home-folder paths, e-mail addresses, LAN addresses, GPS coordinates, and secrets hidden where a text search does not look (bypassed ComfyUI nodes, JSON stored inside a string, compressed PNG chunks, hidden MP4 boxes).
Questions it answers
- Does this PNG, JPEG or MP4 carry a key, a webhook, a user name in a path, an e-mail address, a LAN address or GPS?
- Where exactly: which file, which container, which ComfyUI node, and was that node bypassed (it did not run, but its values are still saved)?
- Is anything hidden in a compressed PNG chunk, in JSON stored inside a string, or in an MP4 box that players and
ffprobedo not show? - Which files could not be scanned at all (damaged, too big, WebP or GIF), so nobody mistakes them for checked?
- After I re-export a copy without the metadata, does a second scan of the copy still find what was found?
How it works
The skill is scan-first. scripts/scan.py only reads the files you give it: it never edits, moves or deletes them, and the one file it writes is the JSON report you name. The report never prints the text it found, only the class, the place, the length and a short keyed hash, so a file's metadata cannot put instructions into a report your agent reads. The skill then tells your agent how to export a copy without the metadata (ffmpeg and Pillow commands, run on synthetic files and labelled per build) and to check the copy by scanning it again; another program makes the copy and the scanner checks it, so the check is not circular. The skill tells the agent to give a fixed report: verdict first, then findings, then how to fix, then what the scan cannot see. A file or folder name can contain text that a shell runs, so the skill tells the agent to paste a file name into a command only if the name is plain (ASCII letters, digits, spaces and . _ ( ) + -), and only if the folder you name has no ", $, backtick, %, ! or typographic double quote in it; for any other name it tells the agent to ask you to copy or rename the file, or move it to a plain-named folder, first.
Honest by design A file the scanner could not scan is not reported as clean: it is listed under COULD NOT FULLY SCAN and the run exits 3 when nothing else was found (1 if another file has a finding). Links and file types it does not read, met inside a folder, are skipped, counted and named under NOTES. A run that scanned no file at all exits 3. No verdict line the skill gives the agent says "clean" or "safe", and it tells the agent never to say a file is fine to share. "Nothing found" is not a guarantee: the known misses are written down (data hidden in pixels or in the compressed image and video streams, the contents of C2PA manifests, secrets that were split or re-encoded, vendor key formats outside its list of eight vendors, webhooks of providers other than Discord and Slack). It does not judge your prompt text or what the picture shows.
It does not replace exiftool or trufflehog. exiftool shows more tags than this scanner reads; trufflehog has far more vendor detectors and can check whether a key is live. What this adds is classification (paths, e-mail addresses, LAN addresses, GPS, hidden MP4 boxes), node and bypassed-node attribution, decoding of compressed PNG chunks, a report that never prints the leaked text, and an exit code that does not read a file it knows it could not scan as clean. In a head-to-head on a synthetic corpus (240 files, 194 planted entries; exiftool 13.59, trufflehog 3.99.2), trufflehog scanned none of the real PNG, JPEG or MP4 files as shipped (the only two files it scanned were a text file and random bytes named .png) and, with workarounds, flagged 27 of 68 planted key entries (the scanner: 68 of 68); exiftool's dumps showed 189 of the 194 entries in clear text and classified none. Much of that gap is the shape of the corpus: counted by underlying key, the report lists 32 OpenAI-legacy keys (trufflehog flagged 1; only 1 of the corpus's 43 OpenAI-legacy values has the marker it needs) and 7 lone AWS key ids it never reports, and excluding OpenAI-legacy values and lone AWS key ids trufflehog flagged 33 of the 40 remaining effective K1 entries (82%), a count on a different basis from the 27 of 68. The scanner matched 194 of 194, but 91 of those 194 are classes trufflehog has no detector for, and the scanner was built against that corpus, so that is not an estimate of real-world recall. The numbers and every caveat are in the package.
What is inside
One scanner (scan.py, standard library only), a demo-file maker (make_demo.py) that writes six synthetic files with invented secrets, five reference notes (what leaks and where, how to export a copy, reading the report, known misses, the comparison), a real sample report in Markdown and JSON, and an example session that includes a run that cannot scan everything. A sixth note (references/no-skill-folder.md) is for an agent that receives the skill as text with no skill folder: it names the one file a scan needs (scan.py, about 145 KB) and gives a short program that writes each file the agent loads into a fresh folder in the system temp folder only when its SHA-256 is the published one, and checks that folder before anything runs.
Requirements
Python 3.9 or newer. ffmpeg is needed only for the export commands and the demo MP4; ffprobe is optional. The 692 automated tests ran on Windows 11 and on Linux in Docker (see the compatibility note). They live in the public repository and are not in the zip. Not tested: macOS, and any agent end to end. Free.
Source and licence Source code and tests: https://github.com/sawizzle/share-safe-renders (MIT OR Apache-2.0). The copy you get through Agensi is under Agensi's buyer terms (a personal, non-transferable licence).
- Does not see data hidden in pixels, or inside the compressed image and video streams, and does not look at what the picture shows. Does not read the contents of C2PA / Content Credentials manifests.
- Opens only PNG, JPEG, MP4/MOV and plain JSON. WebP, GIF, WebM, HEIC/AVIF phone photos, TIFF, PDF and the other formats on its list are reported as not scanned (exit 3 when nothing else was found; with another finding the run exits 1 and still lists them), never as clean. Other file types met while walking a folder are skipped, counted and named under NOTES; the run still exits 0 unless you add --strict.
- Misses secrets that were split, base64- or hex-encoded or glued to a letter, placeholder values (changeme, your_api_key), secret field names outside its list (cookie, credentials, session_id, auth, key), vendor key formats outside the eight vendors it knows, webhook URLs of providers other than Discord and Slack (Teams, Zapier, n8n, Make, Telegram bot URLs), credentials written inside a URL, and a JWT outside a secret-named field.
- Reports only the three private IPv4 ranges and .local/.lan/.internal host names; not IPv6, loopback, link-local, CGNAT or public addresses. Ignores e-mail addresses on example.com, example.org, example.net and example.edu and on the reserved .example, .test, .invalid and .localhost endings; other .org, .net and .edu addresses are reported.
- Does not judge prompt text, model names, seeds or node layout; they travel with the embedded workflow. GPS in a camera's binary telemetry track (GoPro, DJI) is probably missed (no such file was available to test).
- It is a checker. It does not remove metadata; the export commands (ffmpeg, Pillow) were run on synthetic files only, and results differ by build: an ffmpeg PNG re-encode keeps a PNG's EXIF (GPS and e-mail) on ffmpeg 8.1.1 and 9.0.1 but not on 5.1 and 7.1;
-c copyon a JPEG keeps the GPS block; a plain export of an animated PNG fails and leaves a one-frame file; the MP4 command keeps one video and one audio stream; and ffmpeg's -n option keeps an existing output file but exits 0 on 7.1, 8.1.1 and 9.0.1 and 1 on 5.1. A plain Pillow save keeps metadata on every Pillow version tested (a PNG's EXIF block, a JPEG's COM comment or XMP packet, depending on the version), so the documented command passes empty exif, comment and xmp values; it writes only the first frame of an animated file. Always scan the copy again. - The comparison numbers come from a synthetic corpus the scanner was built against (in-sample). trufflehog's custom-detector mode was not tested.
- If your agent receives the skill as text, the route in references/no-skill-folder.md needs the agent to run a short Python program given on standard input and to load each script on its own (scan.py, about 145 KB, for a scan). A file that comes back in the conversation must be written by the agent's own file tool, and the program writes only the published bytes and refuses any copy whose content differs from the published file (it repairs only line ends, a byte order mark and the final newline); then the note tells the agent to stop and tell you. The route was tested on a simulated answer of the current shape and the program on real saved answers, not with every agent or host. Do not use it on a Windows machine whose temp folder, or a folder above it, other users can write.
- Tested on Windows 11 and on Linux x86-64 in Docker, not on macOS. A key that was already shared stays exposed; removing metadata does not take it back.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
Tools used: Terminal / Shell, Read Files, Write Files. NOT used: Browser, Network Access, Environment Variables of their own (the scripts read PATH only, to find ffprobe and ffmpeg in its absolute directories; the current folder is never searched). File scopes: read the files and folders that you name. For a folder, scan.py opens every file in it (links are not followed) and reads its first bytes to tell what it is, whatever the extension; it reads a file fully only if those bytes say PNG, JPEG, MP4/MOV or JSON, or the extension is .json. Write: the single --json report path that you name (plus a temporary file beside it while it is written, `<report>.tmp-` and random characters, made so that it can never be an existing file or a link) and, for make_demo.py, the folder that you name. When your agent has the skill only as text (no skill folder), references/no-skill-folder.md tells the agent itself to load scripts/scan.py (and make_demo.py for the demo) and to write it with a short checking program into a fresh ssr_* folder that the program makes in the system temp folder (for a file shown in the conversation, the note tells the agent to write a scratch file of its own first and to delete it afterwards, and allows writing the file straight into that folder and checking it), and to delete that folder afterwards. Notes: scan.py reads the files you give it and nothing else, opens no network connection, reads no environment variable of its own (it asks the operating system for random numbers: the key of the hash prefix and the name of the temporary file), and never deletes a file. SKILL.md tells the agent to put the report in a folder you name or in a fresh private folder it creates, not loose in the shared temp folder. scan.py refuses to replace an existing --json file that is not empty or an earlier report of its own, and refuses a --json path inside a folder it is scanning. It starts one program, the optional ffprobe, with an argument list (never a shell string) and the file protocol only, and ignores it if absent. make_demo.py starts one program, ffmpeg, with an argument list, only to build the demo MP4; it skips the MP4 with --no-mp4 and refuses to overwrite existing files unless --overwrite is given. The scripts themselves still delete nothing: the temporary folder of an agent that has the skill only as text is the agent's own (references/no-skill-folder.md).