Guide · Agent skill
How to use the akf-trust-metadata agent skill: setup and examples
Updated · written with AI from the project's own documentation and checked against our catalogue data
The akf-trust-metadata skill teaches an AI agent to attach a metadata record to files it creates or changes, recording which agent made the change, what evidence supports it, and a confidentiality label. It is written for teams that need an audit trail of AI-generated or AI-modified files, developers passing files between people or systems, and compliance reviewers. After this guide you will know what an agent skill is, how to install this one, how to ask for its actions in plain language, and what limits to respect.
AKF Trust Metadatasickn33/agentic-awesome-skills · MIT Download the skill Quick start
- 1Install the akf tool. Run pip install akf in a Python environment, or use npm install akf-format. Check the installed CLI help.
- 2Download and unzip the skill. Get the ZIP from the page where you found this guide and unzip it to get the akf-trust-metadata folder.
- 3Place the skill folder. Copy it to ~/.claude/skills/ for personal use or to .claude/skills/ in a project. Alternatively, upload it under Settings → Capabilities → Skills.
- 4Ask for a stamp. Describe the file, the agent that made the change, and what was done, so the agent can run akf stamp with accurate evidence.
- 5Verify the result. Run akf read or akf inspect on the file and confirm the content still looks right.
What is an agent skill and what does this one do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches the skill's description. This skill is named akf-trust-metadata and is based on AKF, described as an AI-native file format that works like EXIF for AI-generated content.
Once installed, the agent can stamp a file it created or edited with trust and provenance metadata, read and inspect existing metadata before changing a file, and run advisory checks against named regulations such as the EU AI Act, HIPAA, SOX, or NIST AI RMF.
Install the akf command-line tool
The skill depends on an external command-line tool called akf. The source gives two options: install the Python package with pip, or use the npm package akf-format as an alternative. Python with pip is required for the pip route.
The skill's own documentation recommends a reviewed, project-pinned installation, so check the installed CLI help before relying on version-specific flags.
- Python route: pip install akf
- npm route (alternative): npm install akf-format
- Project home page: https://akf.dev
- Source code: https://github.com/HMAKT99/AKF
pip install akfInstall the skill in Claude
Download the ZIP for this skill from the page where you found this guide and unzip it. Then place the resulting skill folder in one of two locations. For a personal install available across projects, use ~/.claude/skills/. For a single project, use the .claude/skills/ folder inside that project.
If you use the Claude apps instead, you can upload the skill under Settings → Capabilities → Skills.
- Personal: ~/.claude/skills/
- Project only: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills, then upload
How to trigger the skill with natural requests
You do not run the skill directly. You describe the task, and the agent uses the skill when the request matches its description: stamping, inspecting, or auditing provenance and trust metadata on AI-generated or AI-modified files for compliance, review, or handoff.
Name the file, the agent, and what the agent did. Specific requests give the agent the evidence it needs to write an accurate record.
- "Stamp report.docx with provenance metadata, noting it was generated from my prompt."
- "Check the existing trust metadata on src/billing.py before you change it."
- "Run a quick EU AI Act advisory check on the metadata of these images."
Stamp files and read metadata from the command line
The core action is stamping. After the agent creates or modifies a file, it runs akf stamp with the file path, an agent name, and a short description of the work. Evidence should describe what was actually done, such as "generated from user prompt", "refactored existing code", "tests pass", or "docs reviewed". Only write "tests pass" if the named checks really passed for that exact file.
Before changing an existing file, the agent reads its metadata with akf read and gets detailed scores with akf inspect.
- Stamp: akf stamp <file> --agent <agent-name> --evidence "<what you did>"
- Read existing metadata: akf read <file>
- Detailed trust scores: akf inspect <file>
akf stamp <file> --agent <agent-name> --evidence "<what you did>"Run advisory checks and set confidentiality labels
The skill can run advisory audits against named regulations. These commands inspect metadata fields only. Their output cannot certify regulatory compliance, so verify applicability and requirements separately.
Labels control classification. Use confidential for finance, secret, or internal paths, public for README files, docs, and examples. If you do not set one, the default is internal.
- --label confidential for finance, secret, or internal paths
- --label public for README, docs, and examples
- Default label: internal
akf audit <file> --regulation eu_ai_act
akf audit <file> --regulation hipaa
akf audit <file> --regulation sox
akf audit <file> --regulation nist_aiSafety notes before you use it
Stamping modifies files. Confirm the target scope first, keep originals, and verify both the metadata and the content afterward. The skill is marked with a critical risk level and runs scripts. It does not use the network.
Metadata may travel with a shared file, so keep confidential data out of it. Trust scores and labels are self-declared, not independent proof that code or documents are accurate or legally compliant.
- Keep a copy of originals before stamping
- Do not put confidential details in evidence text if the file will be shared
- Treat audit output as advisory and get expert review for regulatory decisions
What people use it for
- Recording which files in a codebase were AI-generated or AI-modified for a code review trail.
- Labelling documents handed from one team to another with their evidence and confidentiality level.
- Adding provenance metadata to images or PDFs before they are shared.
- Running a quick advisory look at metadata fields before a formal compliance audit.
Frequently asked questions
Is the akf-trust-metadata skill free?
The skill is published under the MIT licence, as listed on this page. The akf tool is installed separately with pip or npm, and this guide does not state any pricing.
What do I need to install the akf-trust-metadata skill?
You need Python with pip to install the akf package, or npm to use the akf-format package instead. You also need the skill folder, which you download as a ZIP and unzip into the Claude skills location.
Where do I put the skill in Claude?
Place the unzipped folder in ~/.claude/skills/ for personal use or in .claude/skills/ inside a project. In the Claude apps you can upload it under Settings → Capabilities → Skills.
Does this skill prove my files are compliant with the EU AI Act or HIPAA?
No. The skill's documentation says the advisory audit output cannot certify regulatory compliance. It inspects metadata fields only, so you should verify requirements and get expert review.
Does the skill need internet access?
The skill is listed as not using the network. It does run scripts and depends on the external akf command-line tool, so review the code before installing.
How is this different from EXIF?
EXIF stores camera data in photos. This skill's source describes AKF as the native metadata format for AI-generated content, storing trust scores, source provenance, and compliance fields across formats such as DOCX, PDF, images, and code.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you need a consistent record of which files an AI agent created or changed, how they were reviewed, and how sensitive they are. Skip it if you need a legal or regulatory sign-off, if you expect the metadata to prove code or documents are accurate, or if you plan to share confidential files with metadata attached.
Download the skill