dingdawg-loop 1.0.0

Governed scheduling for Python agents — CrewAI, LangGraph, and custom agents in 2 lines.

dingdawg-loop

Governed scheduling for Python agents in 2 lines. Works with CrewAI, LangGraph, AutoGen, and any custom agent framework.

from dingdawg_loop import schedule_governed

schedule_governed(agent_id="compliance-scanner", cron="0 9 * * *", verifier="neurosymbolic")

Every execution runs through the DingDawg governance gate before anything fires. Deny = nothing executes. No exceptions.

---

Prerequisites

Node.js >= 18 and the DingDawg Loop MCP server:

npm install -g dingdawg-loop

Then install this package:

pip install dingdawg-loop

---

Use cases

1 — Daily compliance report (CrewAI)

from dingdawg_loop import schedule_governed, execute_governed
from crewai import Crew, Agent, Task

compliance_crew = Crew(agents=[...], tasks=[...])

def run_compliance(loop_id: str):
    compliance_crew.kickoff()

loop = schedule_governed(
    agent_id="compliance-scanner",
    cron="0 9 * * *",           # 09:00 every day
    action_type="generate_report",
    risk_tier="medium",
    description="Daily SOC-2 compliance scan and PDF report generation",
    verifier="neurosymbolic",
    on_execute=run_compliance,
)

print(loop)
# {"loop_id": "abc123", "status": "registered", "schedule": "0 9 * * *"}

2 — Recurring data sync (LangGraph)

from dingdawg_loop import schedule_governed, execute_governed

loop = schedule_governed(
    agent_id="data-sync-agent",
    cron="*/15 * * * *",        # every 15 minutes
    action_type="data_sync",
    risk_tier="low",
    description="Sync CRM records to analytics warehouse",
    verifier="neurosymbolic",
)

# Manually trigger when needed (governance gate runs first)
result = execute_governed(loop["loop_id"], notes="Manual sync before board meeting")
print(result["outcome"])   # "executed" | "denied" | "review_queued" | "error_skipped"

3 — High-risk financial action with inline gate

from dingdawg_loop import GovernanceGate

gate = GovernanceGate(fail_closed=True)

receipt = gate.govern(
    agent_id="finance-agent",
    action_type="wire_transfer",
    description="Transfer $42,000 to vendor account #8821",
    risk_tier="critical",
)

if receipt["decision"] == "allow":
    execute_wire_transfer(...)
else:
    notify_ops_team(receipt)

---

API reference

schedule_governed(...) -> dict

Register a governed loop. Returns {"loop_id", "status", "schedule"}.

Param	Type	Default	Description
agent_id	str	required	Unique agent identifier
cron	str	required	Standard cron expression
action_type	str	"scheduled_execution"	What the agent does
risk_tier	str	"low"	"low" / "medium" / "high" / "critical"
description	str	""	Human-readable description
verifier	str	"neurosymbolic"	"neurosymbolic" or "local"
on_execute	callable	None	Callback fired on governance allow
fail_closed	bool	True	Skip execution if governance unreachable

execute_governed(loop_id, notes="") -> dict

Manually trigger a governed loop. Governance runs first. Returns {"outcome", "receipt_id", "risk_score"}.

Possible outcomes: "executed" / "denied" / "review_queued" / "error_skipped"

list_loops(agent_id=None) -> list[dict]

List registered loops. Optionally filter by agent_id. Degrades gracefully to in-process registry if MCP is unreachable.

GovernanceGate

Inline governance check for one-off actions outside of a scheduled loop.

gate = GovernanceGate(fail_closed=True, timeout=5.0)
receipt = gate.govern(agent_id, action_type, description, risk_tier)
# receipt["decision"] == "allow" | "deny" | "review"

---

Fail-closed behaviour

By default all functions are fail-closed: if the npx dingdawg-loop MCP process is unreachable, times out, or returns an error, execution is skipped and the function returns {"outcome": "error_skipped", "reason": "..."}.

To disable (not recommended for production):

schedule_governed(..., fail_closed=False)
GovernanceGate(fail_closed=False)

---

Environment variables

Variable	Default	Description
DINGDAWG_NPX_CMD	npx	Override the npx binary path
DINGDAWG_LOOP_PKG	dingdawg-loop	Override the npm package name

---

Requirements

• Python >= 3.8
• Node.js >= 18
• npm install -g dingdawg-loop
• Zero Python dependencies (stdlib only)

---

License

MIT — ISG / DingDawg