How to use the CTF Stego Media Analysis agent skill

What is an agent skill, and what does this one do?

An agent skill is a folder with a SKILL.md file that tells an AI agent how to handle a particular kind of task. The agent loads the skill when your request matches its description.

This skill is for hidden data inside media. The source describes it as a CTF-sandbox workflow for image, audio, video, document and container steganography. It helps the agent check file structure first, then rank likely hiding places by evidence, such as alpha or palette channels, least-significant bits, appended trailers, metadata, thumbnails, audio spectrograms or video frames.

• Confirms the real container type, dimensions, duration, codec and chunk layout before guessing
• Checks metadata, thumbnails, sidecar files and appended trailers early
• Ranks candidate channels by evidence instead of brute forcing every transform
• Keeps each extracted layer separate so the steps can be reproduced
• Stops when the hidden payload is reproduced, not just suspected

Check the requirements before you install

This skill is not standalone. An orchestrator skill named ctf-sandbox-orchestrator must already be active, and it must have set the sandbox assumptions and routed your task to this skill. If it has not, the source says to return to ctf-sandbox-orchestrator first.

The skill loads a reference file, references/stego-media.md, which holds the media checklist, channel ranking guide and evidence packaging notes. The workflow implies command-line media and binary inspection tools, for example for headers, EXIF, frame and channel analysis, but the files do not name specific software. You will need to supply those tools yourself.

• Required: the ctf-sandbox-orchestrator skill, already active
• Included: references/stego-media.md, loaded during use
• Not named in the files: the specific inspection tools you will need
• The skill runs no scripts and uses no network

Install the skill from the ZIP download

Download the ZIP for this skill from the page where you found this guide. Unzip it into one of the following locations, depending on how you want to use it.

For a personal setup, unzip the folder into ~/.claude/skills/. For a single project, unzip it into the project's .claude/skills/ folder. The skill folder should contain SKILL.md, agents/openai.yaml and references/stego-media.md.

If you use the Claude apps instead, open Settings, go to Capabilities, then Skills, and upload the skill there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

How do you trigger the skill with natural requests?

You do not need a special command. Describe the problem in plain language and mention the media type and what you suspect. The skill's description lists the kinds of requests it handles, such as inspecting metadata, alpha or palette channels, LSBs, thumbnails, appended trailers, QR fragments, transcoding artifacts, or recovering a hidden payload without blind brute force.

The source says the skill should be used only after $ctf-sandbox-orchestrator has set up the sandbox context. Start that session with the orchestrator first, then ask for media analysis. The skill replies in Simplified Chinese by default unless you explicitly ask for English.

• Good: 'Using the orchestrator context, check this PNG for data in the alpha channel and appended trailers.'
• Good: 'Inspect this WAV file for LSB changes and check the spectrogram for text.'
• Good: 'This MP4 may hide frames. Check frame order and deltas, then tell me which channel holds the payload.'
• Less useful: 'Crack this file.' This asks for blind brute force, which the skill is designed to avoid.

What the agent does step by step

The skill works in three stages. First it establishes what the container really is, using headers, chunk tables, metadata, container indexes, thumbnails and file size anomalies. It compares the declared format with the observed structure to catch polyglots, appended archives or malformed trailers.

Next it inspects candidate channels, such as alpha, palette order, colour planes, LSBs, spectrogram features, document object streams or video frame deltas. Finally it rebuilds the payload path in order: container, channel or carrier, extraction, decompression or decode, then final parse. If the problem becomes mainly cryptographic after extraction, the skill hands off to a broader crypto skill.

• Record exact offsets, frame numbers and channel boundaries that look promising
• Note whether the payload is plain bytes, another media layer, compressed data or an encrypted blob
• Keep each extracted stage as a separate file and record the command used to produce it

What should you expect in the output?

The skill asks the agent to keep a compact evidence record. This includes file structure facts such as offsets, chunks, frame numbers, stream names, metadata keys and trailer size. It also includes intermediate extractions and the exact transform that produced each one.

When the payload is recovered, the final answer should name the payload and the channel that produced it. The reference file also asks the agent to say clearly whether the result is plaintext, another container, compressed data or ciphertext.

• File structure facts: offsets, chunks, frame numbers, stream names, metadata keys, trailer size
• Intermediate extractions and their exact commands
• The final payload and the channel that produced it

Safety notes and limits

The skill runs no scripts and uses no network access. The source says it is intended for recovering payloads in a controlled competition setting, inside an isolated sandbox. It is not meant for hiding or exfiltrating data for evasion or malicious purposes.

Do not use it on real user media on production systems without authorisation. The workflow involves running extraction tools against files of unknown origin, so work in an isolated environment. Pure cryptography problems belong to a separate crypto skill.

• Use it on CTF challenges or files you are authorised to analyse
• Work in an isolated sandbox when handling files of unknown origin
• Avoid blind brute force; the skill is built around evidence-first checks
• Send pure crypto questions to a crypto skill instead