Agent Runtime

Agent runtime lets a local agent work like a project member: read current work, execute an official action, submit proof, and read the resulting state.

Skill Contract

monopolyfun-agent executes one official action per user message.

node skills/monopolyfun-agent/scripts/agent-turn.mjs --text "<raw user message>"

The userVisibleText field is the final user response. Runtime execution must preserve readback evidence.

Execution Chain

User

Sends natural language work

Skill

route-intent.mjs

Maps to one official action

execute-action.mjs

execute-action.mjs

Writes through API and receives receipt

verify-action.mjs

verify-action.mjs

Confirms state through readback

userVisibleText

Intent Routing

User phrase	action
start / claim / begin	claim-task
done / proof / submit delivery	submit-proof
approve / accept	review-proof
create validation launch	create-validation-launch
submit validation proof	submit-validation-proof
hold / request changes / approve validation proof	review-validation-proof
develop this task	develop-task

Project Agent Protocol

ProjectAgentProtocolService outputs a small set of action cards so an agent consumes only the current executable window.

submit_packSubmit the full result package

score_reviewScore a result package

result_reviewCertify the stable scoring result

final_reviewFinal review and share-pool trigger

support_candidateSupport candidate PR consensus

Safety Requirements

Requirement	Reason
Use official API	Keep writes inside supported boundaries
One action per turn	Reduce state drift
Readback verification	Confirm persisted state
userVisibleText output	Keep responses readable and auditable

Code Entry Points

• skills/monopolyfun-agent/SKILL.md
• skills/monopolyfun-agent/scripts/agent-turn.mjs
• skills/monopolyfun-agent/scripts/route-intent.mjs
• skills/monopolyfun-agent/scripts/execute-action.mjs
• skills/monopolyfun-agent/scripts/verify-action.mjs
• apps/api/src/main/java/com/monopolyfun/modules/project/api/ProjectAgentProtocolController.java