AKF helps Openclaw Skills stamp, read, verify, and audit trust metadata in AI-generated files so provenance, confidence, and compliance travel with the content.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install akf
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install akf using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is AKF — Agent Knowledge Format?
AKF (Agent Knowledge Format) is a trust metadata standard for AI-generated content—think EXIF for AI. It helps Openclaw Skills stamp who made a file, what evidence supports each claim, how confident the agent is, and whether a human reviewed it.
By embedding trust data directly into DOCX, PDF, images, code, and memory files, AKF makes every artifact easier to verify, reuse, and audit. For Openclaw Skills workflows, it creates a practical trust layer for agent output, supply-chain verification, and regulated content pipelines.
AKF — Agent Knowledge Format Use Cases
- Before reusing an existing file, run
akf checkto detectOK,STALE,LOW, orUNSTAMPEDcontent. - After an agent creates or edits a document, code file, image, or PDF, stamp it with provenance and evidence.
- Audit output directories for trust gaps, stale claims, and compliance issues.
- Embed trust metadata into native file formats and extract it later for inspection.
- Classify sensitive work with
public,internal,confidential, orrestrictedlabels. - Store knowledge or memory files with trust decay so old facts automatically lose confidence over time.
- Verify downloaded skills or third-party artifacts before loading them into an Openclaw Skills pipeline.
- Re-run evidence with a replay recipe when a claim needs confirmation instead of blind trust.
How AKF — Agent Knowledge Format Works
- Check the file first with
akf checkto determine whether the artifact is fresh, stale, low-trust, or completely unstamped. - Stamp any new or modified output with the producing agent, evidence, optional sensitivity label, and any replay recipe needed for later verification.
- Embed the metadata directly into the file format when supported, so trust information travels with the document instead of living in a separate database.
- Read, inspect, or compute effective trust with
akf read,akf inspect, andakf trustto see the score, provenance tier, and acceptance decision. - Scan directories and audit files to find trust gaps, compliance risks, and aging memory entries whose trust has decayed below threshold.
- Replay evidence when a claim needs falsifiable verification, preserving provenance across Openclaw Skills workflows and chained file reuse.
AKF — Agent Knowledge Format Setup
- Install the AKF CLI or SDK depending on your workflow.
pip install akf
npm install akf-format
- Confirm the binary is available and check a file before trusting it.
akf --help
akf check path/to/file
- Stamp every generated or modified artifact in your Openclaw Skills pipeline.
akf stamp output.py --agent openclaw --evidence "generated from user request"
- Add sensitivity labels and stronger evidence for regulated content.
akf stamp report.pdf --agent openclaw --label confidential --evidence "quarterly data from ERP"
- When consuming third-party skills, memory files, or downloaded assets, verify them first.
akf check downloaded-skill.md
akf check memory/facts.md
AKF — Agent Knowledge Format Data Schema & Taxonomy
AKF organizes trust around the file itself, using embedded metadata plus CLI-readable inspection output. It does not depend on a separate trust database for core verification.
Metadata placement
| Layer | What it stores | Purpose |
|---|---|---|
| File frontmatter | akf block for markdown-based assets such as SKILL.md and memory files |
Human-readable provenance and trust claims |
| Native embedded metadata | Trust records embedded inside DOCX, PDF, images, and supported code workflows | Keeps trust attached to the artifact |
| CLI diagnostics | Output from akf check, akf read, akf inspect, akf trust, akf scan, and akf audit |
Fast verification and compliance review |
Core fields
| Field | Meaning | Example |
|---|---|---|
akf.v |
AKF schema version | 1.0 |
akf.id |
Unique issuance identifier | akf-742cec9663a9 |
akf.agent |
Producing agent | claude-code |
akf.at |
Issuance timestamp | 2026-07-14T13:32:23Z |
akf.label |
Sensitivity class | public, internal, confidential, restricted |
akf.hash |
Content hash for integrity checks | sha256:2e5ff5b05a9232a2 |
akf.inherit |
Whether provenance chains forward across derived files | true |
akf.ext |
Extension support flag | false |
akf.sv |
Skill-schema version | 1.1 |
Claim schema
| Field | Meaning | Notes |
|---|---|---|
claims[].c |
Claim text | The statement being trusted |
claims[].t |
Trust score from 0.0 to 1.0 |
Drives ACCEPT / REVIEW / REJECT decisions |
claims[].id |
Claim identifier | Stable reference for audits |
claims[].src |
Source attribution | May be unspecified or other provenance source |
claims[].tier |
Provenance tier | Higher tiers indicate stronger evidence |
claims[].ver |
Verified flag | true or false |
claims[].ai |
AI involvement flag | Indicates whether AI contributed |
claims[].decay |
Trust half-life in days | Used for aging and memory workflows |
claims[].kind |
Artifact type | Examples: skill, memory |
claims[].evidence |
Evidence records | Human review, tests, or replay output |
Trust taxonomy
public: README, docs, and open-source examplesinternal: Default general work outputconfidential: Finance, legal, medical, and HR contentrestricted: Credentials, secrets, and PII
Decision model
| Score range | Decision | Meaning |
|---|---|---|
0.80-1.00 |
ACCEPT | High confidence and well-evidenced |
0.50-0.79 |
REVIEW | Moderate confidence, needs verification |
0.00-0.49 |
REJECT | Low confidence or unreliable |
Openclaw Skills uses this schema to keep trust attached to the artifact, not just to a prompt or chat transcript.
AKF — Agent Knowledge Format Advanced Features
- Falsifiable evidence with replay recipes: stamp claims with
--replay, then useakf replayto confirm, drift-detect, or refute them. - Native embed/extract workflows for supported file formats, including DOCX and PDF.
- Directory-wide trust scanning and compliance auditing for EU AI Act, HIPAA, SOX, and NIST-oriented workflows.
- Trust decay for memory files so stale knowledge automatically drops below the trust threshold over time.
- Provenance chaining across reused files so Openclaw Skills can preserve trust through multi-step workflows.
- Fast trust gating with one-line
akf checkbefore reusing any file, skill, or downloaded artifact. - Sensitivity labeling plus security scanning across multiple AI content risk classes.
- High-signal trust metadata for Openclaw Skills teams that need auditability without slowing down agent output.
SKILL.md
Loading
Related Openclaw Skills
Tristan RFQ Overseer
Tristan RFQ Overseer is an Openclaw Skills RFQ pipeline that captures requests, prices them transparently, compares suppliers, and stages approved quotes for sending.
Ditto Profile Loader
Loads a user's mined Ditto profile so Openclaw Skills agents follow their real working style instead of starting from zero.
MCP Precheck
Openclaw Skills for mcp-precheck gives you deterministic, registry-backed MCP server vetting before any connection, config edit, or hook install.
Journal Metadata Lookup
Journal Metadata Lookup is an Openclaw Skills utility that resolves a paper identifier into authoritative authorship, venue, and citation metadata in one command.
AIPOU Farming
AIPOU Farming turns completed AI-assisted work into privacy-preserving receipts and optional Base rewards through Openclaw Skills.
Obsidian Vault Cleaner
An Openclaw Skills utility that audits Obsidian and Markdown vaults for duplicates, orphan notes, broken links, and unused attachments, then recommends safe fixes.
Featured*
