{ "markdown": "---\n_akf: '{\"v\":\"1.0\",\"claims\":[{\"c\":\"Trust metadata for README.md\",\"t\":0.7,\"id\":\"1979cbeb\",\"src\":\"unspecified\",\"tier\":5,\"ver\":false,\"ai\":true,\"evidence\":[{\"type\":\"other\",\"detail\":\"updated certify and github action references\",\"at\":\"2026-03-18T04:21:48.869226+00:00\"}]}],\"id\":\"akf-c33254656fc5\",\"agent\":\"claude-code\",\"at\":\"2026-03-18T04:21:48.870623+00:00\",\"label\":\"public\",\"inherit\":true,\"ext\":false,\"sv\":\"1.1\"}'\n---\n

\n \"AKF\n

\n\n

\n \"Stars\"\n \n \n \n \"AKF\n \n \n \n

\n\n

AKF — The AI Native File Format

\n\n

\n A stamp costs ~15 tokens. Re-verifying costs 15,000.
\n Think EXIF for AI — trust metadata that travels with every AI-generated file,
\n so the next agent (or human) knows what it can build on.\n

\n\n

\n \"AKF\n

\n\n

\n ⭐ Star on GitHub ·\n Website ·\n 🛡️ Is This AI-Generated? ·\n 🇪🇺 Compliance Check ·\n 📊 Repo Scanner\n

\n\n

\n Quickstart ·\n AI Agents ·\n MCP Server ·\n Agent Skills ·\n CLI ·\n Website\n

\n\n---\n\n## Paste this to your agent\n\nDon't install AKF yourself — let your agent do it. Paste this into Claude Code, Cursor, OpenClaw, or any coding agent:\n\n```\nInstall AKF (pip install akf, or pipx install akf) and set it up: run `akf init --git-hooks`.\nFrom now on:\n- After you create or modify a file, stamp it:\n akf stamp --agent --evidence \"\"\n- Before building on any existing file, check it first:\n akf check \n OK = fresh stamp with verified evidence, skip re-verification. STALE/LOW/UNSTAMPED = verify before trusting.\nIf you support MCP, also register the server: {\"mcpServers\":{\"akf\":{\"command\":\"python\",\"args\":[\"-m\",\"mcp_server_akf\"]}}}\n```\n\nYour agent installs its own provenance layer — and every file it touches carries trust metadata from then on.\n\n## Why agents need this\n\nAgents redo work constantly: re-read files, re-run tests, re-derive conclusions — because nothing tells them what was already verified. A stamp is cached verification state:\n\n```console\n# Session 1 — agent fixes auth, tests pass\n$ akf stamp auth.py --agent claude-code --evidence \"42/42 tests passed\"\n\n# Session 2 — tomorrow, any agent, any tool\n$ akf check auth.py\nOK trust=0.65 agent=claude-code evidence=test_pass age=1d claims=1\n# → build on it, skip re-verification\n\n# Someone edits auth.py without re-testing\n$ akf check auth.py\nSTALE trust=0.65 agent=claude-code evidence=test_pass age=1d claims=1 reason=modified_after_stamp\n# → re-verify before trusting (exit code 1 — gate CI or hooks on it)\n```\n\nStamps are trail markers agents leave for other agents — across sessions, across tools (Claude Code → Cursor → Copilot), across teams. Humans get the same trail: who made this file, AI or human, was it tested, can we trust it.\n\n| What travels with the file | How |\n|------|-----|\n| **Trust score** | 0–1 confidence, weighted by evidence and source tier |\n| **Verification evidence** | tests passed, type check clean, human reviewed — with timestamps |\n| **Source provenance** | SEC filing → analyst → AI agent chain |\n| **Compliance** | One command: `akf audit file --regulation eu_ai_act` |\n\n## Don't trust the stamp — re-run it\n\nA signature proves *who* said it; a replay proves it *could have been true*. A stamp can carry a **falsifiable probe recipe**, so the next agent re-verifies the claim instead of trusting the label:\n\n```console\n# Stamp with a recipe that can be re-run\n$ akf stamp auth.py --agent claude-code --evidence \"42/42 tests passed\" --replay \"pytest -q\"\n\n# Later — re-run the probe instead of trusting the stamp\n$ akf replay auth.py --run\nCONFIRMED inputs=intact\n\n# A dependency moved since the stamp was written\n$ akf replay auth.py --run\nCONFIRMED_DRIFTED inputs=drifted # probe still passes, but against a changed world — re-check\n```\n\n`REFUTED` when the probe fails, `UNREPLAYABLE` when there's no recipe. This is the answer to \"a trusted source can still be wrong\": trust stops depending on who signed it, for any claim with a runnable check.\n\n## Quickstart\n\n```bash\npip install akf # Python\nnpm install akf-format # TypeScript / Node.js\n\nakf doctor # Check your install — detects PATH issues and guides setup\n```\n\n> **`akf` command not found?** Run `akf doctor` to auto-detect your setup, or use `python3 -m akf` (always works).\n> - Install with pipx: `pipx install akf` (recommended — auto-handles PATH)\n> - **Windows:** use `python3 -m akf` or install via `pipx`\n\n```bash\n# The core loop — stamp what you verified, check before you trust\nakf stamp auth.py --agent claude-code --evidence \"42/42 tests passed\"\nakf check auth.py # OK trust=0.65 agent=claude-code evidence=test_pass age=0d\n```\n\n```python\nimport akf\n\n# Same loop from Python\nakf.stamp_file(\"auth.py\", agent=\"claude-code\", evidence=[\"42/42 tests passed\"])\nresult = akf.check_file(\"auth.py\")\nprint(result.summary_line()) # OK trust=0.65 agent=claude-code evidence=test_pass age=0d claims=1\n\n# Embed into Office docs, PDFs, images — any format\nakf.embed(\"report.docx\", claims=[...], classification=\"confidential\")\n\n# Audit for compliance (EU AI Act, HIPAA, SOX, GDPR, NIST AI, ISO 42001)\nresult = akf.audit(\"report.akf\", regulation=\"eu_ai_act\")\nprint(f\"Compliant: {result.compliant}\")\n```\n\n**TypeScript / Node.js** (`akf-format`):\n\n```typescript\nimport { create, validate, effectiveTrust, stampFile } from 'akf-format';\n\n// Create a trust-stamped unit from any AI output\nconst unit = create('Revenue was $4.2B, up 12% YoY', 0.98, {\n source: 'SEC 10-Q',\n agent: 'claude-code',\n});\n\n// Validate against the AKF schema\nconst { valid } = validate(unit);\n\n// Compute effective trust for a claim\nconst trust = effectiveTrust(unit.claims[0]);\nconsole.log(`valid: ${valid}, score: ${trust.score}, decision: ${trust.decision}`);\n\n// Stamp trust metadata directly into a file (markdown, json, code, …)\nstampFile('report.md', { agent: 'claude-code', evidence: 'tests pass' });\n```\n\n> Full TypeScript API and more examples: [`typescript/README.md`](typescript/README.md).\n\n## For AI Agents\n\nAKF is designed **agent-first**. One-line APIs for checking, stamping, streaming, and auditing.\n\n```python\nimport akf\n\n# Check before you trust — can I build on this file without re-verifying?\nresult = akf.check_file(\"auth.py\")\nif result.status == \"OK\": # fresh stamp, verified evidence\n ... # skip re-verification, save the tokens\n# LOW / STALE / UNSTAMPED → verify before trusting\n\n# Stamp with evidence (auto-detected: test_pass, type_check, human_review, etc.)\nakf.stamp(\"Fixed auth bypass\", kind=\"code_change\",\n evidence=[\"42/42 tests passed\", \"mypy: 0 errors\"],\n agent=\"claude-code\", model=\"claude-sonnet-4-20250514\")\n\n# Stream trust metadata in real-time\nwith akf.stream(\"output.md\", model=\"gpt-4o\") as s:\n for chunk in llm_response:\n s.write(chunk)\n\n# Trust-annotated git commits (uses git notes)\nakf.stamp_commit(content=\"Refactored auth module\", kind=\"code_change\",\n evidence=[\"all tests pass\"], agent=\"claude-code\")\nprint(akf.trust_log(n=10)) # + ACCEPT ~ LOW - REJECT ? none\n```\n\n## Multi-Agent Teams\n\nAKF supports multi-agent orchestration — Claude Agent Teams, Copilot Cowork, Codex multi-agent, and any A2A-compatible platform.\n\n```python\nimport akf\n\n# Agent-to-agent delegation with trust ceiling\npolicy = akf.DelegationPolicy(\n delegator=\"lead-agent\", delegate=\"research-bot\",\n trust_ceiling=0.7, allowed_actions=[\"search\", \"summarize\"]\n)\nresult = akf.delegate(parent_unit, policy)\n\n# Multi-agent streaming session\nwith akf.TeamStream([\"research\", \"writer\", \"reviewer\"]) as ts:\n ts.write(\"research\", \"Found 3 sources\", confidence=0.8)\n ts.write(\"writer\", \"Drafted summary\", confidence=0.75)\n ts.write(\"reviewer\", \"Approved with edits\", confidence=0.9)\n scores = ts.aggregate() # per-agent + team trust\n\n# Cross-platform agent identity\ncard = akf.create_agent_card(name=\"Research Bot\", platform=\"claude-code\",\n capabilities=[\"search\", \"summarize\"])\nakf.verify_agent_card(card) # SHA-256 hash verification\n\n# Team certification (per-agent breakdown)\nreport = akf.certify_team(\"src/\", min_trust=0.7)\n# report.all_agents_certified — each agent must individually pass\n```\n\n**CLI:**\n```bash\nakf agent create --name \"Bot\" --platform claude-code --capabilities search,summarize\nakf agent list\nakf agent verify \nakf agent export-a2a --output card.json # A2A protocol bridge\nakf agent import-a2a card.json\nakf certify src/ --team # Per-agent breakdown\n```\n\n## MCP Server\n\nAKF ships an [MCP](https://modelcontextprotocol.io) server so any AI agent can create, validate, scan, and audit trust metadata.\n\n```bash\n# Install from the repo\npip install ./packages/mcp-server-akf\n```\n\n```json\n{\n \"mcpServers\": {\n \"akf\": {\n \"command\": \"python\",\n \"args\": [\"-m\", \"mcp_server_akf\"]\n }\n }\n}\n```\n\n**11 MCP tools:** `check_file` · `replay_file` · `create_claim` · `validate_file` · `scan_file` · `trust_score` · `stamp_file` · `audit_file` · `embed_file` · `extract_file` · `detect_threats`\n\n## Ambient Trust\n\nAKF works where AI agents work. Drop a config file, and every AI-generated file carries trust metadata automatically.\n\n| Agent | How it works |\n|-------|-------------|\n| **Claude Code** | Plugin: `/plugin marketplace add HMAKT99/AKF` → `/plugin install akf` — auto-stamp hook + check skill. Or reads `CLAUDE.md` |\n| **Cursor** | Reads `.cursorrules` — stamps AI edits before you review |\n| **Windsurf** | Reads `.windsurfrules` — stamps AI edits with trust metadata |\n| **GitHub Copilot** | Reads `.github/copilot-instructions.md` (native) + shell hook for CLI |\n| **OpenAI Codex** | Reads `AGENTS.md` — stamps files in cloud sandbox and local |\n| **OpenClaw** | Skill on ClawHub: `clawhub install akf` — check/stamp protocol + memory trust |\n| **Hermes Agent** | agentskills.io skill: `hermes skills tap add HMAKT99/AKF` — files, memories, and skill supply-chain |\n| **Manus / Other Agents** | MCP server + shell hook — works with any agent that supports MCP or CLI |\n| **Any MCP agent** | 11 MCP tools — check, replay, stamp, audit, embed, extract, detect, validate, scan, trust, create |\n| **Any CLI tool** | `eval \"$(akf shell-hook)\"` — intercepts `claude`, `chatgpt`, `aider`, `openclaw`, `ollama`, `manus` |\n\n**The trust pipeline:**\n```\nAgent writes code → Git commit stamped → CI runs akf certify → Team reviews with context\n```\n\nSet up in 60 seconds:\n```bash\n# 1. Agent stamps its own work (already in this repo)\ncat CLAUDE.md # or .cursorrules / .windsurfrules / AGENTS.md / .github/copilot-instructions.md\n\n# 2. Git hooks stamp every commit\nakf init --git-hooks\n\n# 3. CI certifies trust on every PR\n# uses: HMAKT99/AKF/extensions/github-action@main\n\n# 4. Shell hook intercepts AI CLI tools\neval \"$(akf shell-hook)\"\n```\n\n## Skills\n\nAKF provides [agent skill files](skills/) that AI agents can discover and use. Drop these into your agent's context:\n\n| Skill | What it does |\n|-------|-------------|\n| [`check.md`](skills/check.md) | Check a file's trust before building on it |\n| [`stamp.md`](skills/stamp.md) | Stamp trust metadata onto AI outputs |\n| [`audit.md`](skills/audit.md) | Audit files for regulatory compliance |\n| [`scan.md`](skills/scan.md) | Security scan files and directories |\n| [`embed.md`](skills/embed.md) | Embed trust metadata into Office/PDF/images |\n| [`detect.md`](skills/detect.md) | Run 10 security detection classes |\n| [`stream.md`](skills/stream.md) | Stream trust metadata in real-time |\n| [`git.md`](skills/git.md) | Trust-annotated git workflows |\n| [`convert.md`](skills/convert.md) | Convert between formats |\n| `delegate` | Agent-to-agent trust delegation |\n| `team` | Multi-agent streaming sessions |\n\n## Format at a Glance\n\n**Compact** (~15 tokens — optimized for AI):\n```json\n{\"v\":\"1.0\",\"claims\":[{\"c\":\"Revenue was $4.2B\",\"t\":0.98,\"src\":\"SEC 10-Q\"}]}\n```\n\n**Descriptive** (human-readable — same data):\n```json\n{\"version\":\"1.0\",\"claims\":[{\"content\":\"Revenue was $4.2B\",\"confidence\":0.98,\"source\":\"SEC 10-Q\"}]}\n```\n\n**Full** (with provenance, decay, AI flags, security):\n```json\n{\"v\":\"1.0\",\"by\":\"sarah@acme.com\",\"label\":\"confidential\",\"inherit\":true,\n \"claims\":[\n {\"c\":\"Revenue $4.2B\",\"t\":0.98,\"src\":\"SEC 10-Q\",\"tier\":1,\"ver\":true,\"decay\":90},\n {\"c\":\"H2 will accelerate\",\"t\":0.63,\"tier\":5,\"ai\":true,\"risk\":\"AI inference\"}\n ],\n \"prov\":[\n {\"hop\":0,\"by\":\"sarah@acme.com\",\"do\":\"created\",\"at\":\"2025-07-15T09:30:00Z\"},\n {\"hop\":1,\"by\":\"copilot-agent\",\"do\":\"enriched\",\"at\":\"2025-07-15T10:15:00Z\"}\n ]}\n```\n\n## Works With Every Format\n\nAKF embeds natively — no sidecars needed for most formats:\n\n| Format | How It Works |\n|--------|-------------|\n| `.akf` | Native standalone knowledge file |\n| `.docx` `.xlsx` `.pptx` | OOXML custom XML part |\n| `.pdf` | PDF metadata stream |\n| `.html` | JSON-LD `