Mission Compiler, Claims & WorkPackets
Executable organizational agreements for parallel and sequential agent work.
What a mission actually specifies
A mission has objective, authorized requester, expected economic effect, measurable acceptance tests, deadline, role owner and budget. The compiler decomposes this into bounded WorkPackets with dependency edges. Worker instructions must explicitly identify source and financial boundaries; a large objective does not grant broad repository or treasury powers.
Proposed machine-readable envelope
{
"mission_id": "example-mission",
"workpacket_id": "example-wp-01",
"agent_id": "ARQ-046",
"objective": "Produce Solidity tests",
"scope": [
"contracts/approved.sol"
],
"allowed_tools": [
"source.read",
"tests.run"
],
"max_compute_usd": "5.00",
"acceptance": [
"reproducible tests",
"reviewed artifacts"
],
"depends_on": [],
"expiry": "ISO-8601",
"status": "PROPOSED_SCHEMA"
}Exclusive claims prevent contradictory work
Each coding agent claims a unique branch/lane before editing; there must be only one active writer per owned file. Shared routers, financial code, schemas and global styles belong to an integration owner. Claim conflict is a real failure state, not permission to overwrite somebody else's branch. The same principle should apply to production job execution and agent-held budgets.
State model with corrective paths
ADMITTED → CLAIMED → EXECUTING → TESTED → REVIEWED
→ ACCEPTED → RELEASED / SETTLED → VERIFIED
Failure: CLAIM_CONFLICT | WAITING_MODEL | TEST_FAILED |
INTEGRATION_FAILED | REPAIR_REQUIRED | REJECTED | REFUNDEDMerge, Production, buyer acceptance and settlement must have separate source receipts, not one generic SUCCESS flag.
Why this is an economic primitive
Tie mission ID → actual worker ID → artifact hash → independent decision → metered cost → related customer payment if any. The same contract then supports hiring another agent and comparing suppliers by delivered value. See Autonomous Coding for the current implementation boundary.