TaskHawk Systems

TaskHawk Systems Governance API

Public governance operations: verify, attest, bind, verify-outcome, and bundle. Successful operations return endpoint-specific evidence; provenance-capable operations append audit records.

Documentation GitHub OpenAPI

Operations 8

Each operation below carries the questions people ask an LLM about it and the instructions they give an agent to run it. Generated by API Evangelist overlay

POST /governance/verify Check a proposed agent action against policy · Verify Action #

Ask an LLM

“How do I check whether an agent's proposed action is allowed by policy before it runs?”

“Can I get a signed allow, constrain or deny decision for a transaction an agent wants to make?”

Tell an agent

Verify a {action_type} action from agent {agent_id} with payload {action_payload} against policy.

Check agent {agent_id}'s {action_type} action {action_payload} using policy template {template_id}.

POST /governance/attest Record a provenance attestation for an agent action · Attest Action #

Ask an LLM

“How do I log what my agent did into a tamper-evident, hash-chained record?”

“Can a third party independently verify an attestation of an agent's action?”

Tell an agent

Attest that agent {agent_id} did {action_description} with details {action_payload}.

Add an attestation for agent {agent_id}: {action_description}, payload {action_payload}, chained to {prior_attestation_hash}.

POST /governance/bind Bind a declared intent to an agent command · Bind Intent #

Ask an LLM

“How can I prove a command was issued in service of a specific declared intent?”

“Can I nest an intent under a parent intent to build a hierarchy?”

Tell an agent

Bind agent {agent_id}'s {intent_type} intent "{intent_description}" to command {command_payload}.

Declare intent {intent_description} of type {intent_type} for agent {agent_id}, binding command {command_payload} with goal {goal_state}.

POST /governance/verify-outcome Confirm an executed action met its declared intent · Verify Outcome #

Ask an LLM

“How do I check that an action my agent already executed actually achieved its intent?”

“Can I allow some tolerance when comparing the end state to the goal state?”

Tell an agent

Verify agent {agent_id}'s outcome for intent {intent_id} and binding {binding_id} given resulting state {actual_state}.

Check whether state {actual_state} satisfies intent {intent_id} (binding {binding_id}, agent {agent_id}) within tolerance {tolerance}.

POST /governance/bundle Generate a compliance evidence bundle for an agent · Generate Bundle #

Ask an LLM

“How do I export certifier-grade compliance evidence for one of my agents?”

“Can I limit an evidence bundle to a specific time range?”

Tell an agent

Generate a compliance evidence bundle for agent {agent_id}.

Build an evidence bundle for agent {agent_id} from {time_range_start} to {time_range_end}.

POST /governance/mpp/session Open a governed streaming payment session · Create a governed MPP payment session #

Ask an LLM

“How do I let an agent stream payments within a fixed budget and time window?”

“Can I cap how fast an agent spends per minute during a payment session?”

Tell an agent

Open a payment session for agent {agent_id} with a budget of {session_budget_cents} cents for {duration_seconds} seconds.

Start an MPP session for {agent_id}: {session_budget_cents} cents over {duration_seconds}s on the {payment_rail} rail.

POST /governance/mpp/heartbeat Report mid-session spend for a drift check · Mid-session drift check #

Ask an LLM

“How do I check an active payment session for budget overruns or unauthorized services?”

“Is there a charge for sending periodic heartbeats during a streaming payment session?”

Tell an agent

Heartbeat session {session_id}: {agent_id} spent {spent_cents}c in {transactions_count} txns on {current_service} at {spending_rate_cents_per_minute}c/min.

Run a drift check on open session {session_id} for agent {agent_id}, now at {spent_cents} cents.

POST /governance/mpp/close Close a streaming payment session · Close an MPP session #

Ask an LLM

“How do I end a streaming payment session and seal its provenance record?”

“Can I record why a payment session ended, like budget exhausted or drift detected?”

Tell an agent destructive · confirm first

Close session {session_id} for agent {agent_id} with final spend {final_spent_cents} cents across {final_transactions_count} transactions.

End payment session {session_id} with reason {close_reason}.

TaskHawk Systems Governance API

Operations 8

Documentation

Specifications

Other Resources

Work with this as data

MCP server

Tools for apis

Call it yourself

Get an API key

OpenAPI Specification