Standard: AISS v2.0 · Full stack: piqrypt.com
Verifiable AI Agent Memory_Cryptographic audit trail for OpenClaw autonomous agents.
Every reasoning step, tool execution (bash, Python, file ops), and task result — signed, hash-chained, tamper-proof.
pip install piqrypt-openclawOpenClaw has OS-level access — it can run bash, write files, execute Python. When an autonomous agent has this much power, you need to know exactly what it did and when. PiQrypt makes every action cryptographically irrefutable.
from openclaw import Agent
from piqrypt_openclaw import AuditableOpenClaw
# Wrap your existing OpenClaw agent
base_agent = Agent(config)
agent = AuditableOpenClaw(base_agent, identity_file="openclaw.json")
# Execute tasks — every step is stamped automatically
result = agent.execute_task(task)
# Export tamper-proof audit trail
agent.export_audit("openclaw-audit.json")
# $ piqrypt verify openclaw-audit.json# Stamp each tool call individually for full traceability
agent = AuditableOpenClaw(base_agent, identity_file="openclaw.json")
# After planning
prev = agent.stamp_reasoning(task.description, plan, model="llama-3.2")
# After each tool call
prev = agent.stamp_tool_call("bash", "ls -la /reports", output, prev)
prev = agent.stamp_tool_call("python", analysis_code, result, prev)
prev = agent.stamp_tool_call("file_write", "report.pdf", "written", prev)# Detect suspicious bash patterns (rm -rf, curl | bash, etc.)
suspicious = agent.get_suspicious_events()
for event in suspicious:
print(f"⚠️ Suspicious action stamped:")
print(f" Tool: {event['payload']['tool']}")
print(f" Hash: {event['payload']['input_hash'][:16]}...")
print(f" Timestamp: {event['timestamp']}")Suspicious patterns are stamped and flagged, not blocked — the audit trail proves what happened.
from piqrypt_openclaw import stamp_action
@stamp_action("file_analysis", identity_file="my-agent.json")
def analyze_sales_data(path: str) -> dict:
return your_analysis_logic(path)
@stamp_action("report_generation", identity_file="my-agent.json")
def generate_report(data: dict) -> str:
return your_report_logic(data)| Event | When |
|---|---|
agent_initialized |
Agent creation |
task_start |
Before task execution |
task_reasoning |
After LLM planning phase |
tool_execution |
Each tool call (bash, Python, file ops) |
task_complete |
After task finishes |
task_failed |
On exception (with error hash) |
All events are Ed25519-signed, SHA-256 hash-chained.
Raw commands and outputs are never stored — only their SHA-256 hashes.
piqrypt verify openclaw-audit.json
# ✅ Chain integrity verified — 24 events, 0 forks
piqrypt search --type tool_execution --limit 10
# Lists last 10 tool calls with timestamps→ docs/OPENCLAW_INTEGRATION.md
- PiQrypt core: github.com/piqrypt/piqrypt
- Issues: github.com/piqrypt/piqrypt/issues
- Support: [email protected]