[{"data":1,"prerenderedAt":240},["ShallowReactive",2],{"docs-\u002Fdocs\u002Fai-roles":3},{"id":4,"title":5,"body":6,"description":232,"extension":233,"meta":234,"navigation":235,"path":236,"seo":237,"stem":238,"__hash__":239},"docs\u002Fdocs\u002Fai-roles.md","AI agent roles",{"type":7,"value":8,"toc":221},"minimark",[9,14,18,21,25,28,34,58,61,65,68,74,80,100,103,106,110,113,117,131,134,138,141,145,159,162,166,169,173,187,190,194,197,201,218],[10,11,13],"h2",{"id":12},"overview","Overview",[15,16,17],"p",{},"FFF is built on a team of specialised AI agents. Each agent has a narrow, well-defined scope. Narrow scope means narrow failure modes - when an agent behaves poorly, the problem is easy to diagnose and the fix is a targeted change to that agent's instructions.",[15,19,20],{},"No agent has end-to-end ownership of a ticket. Handoffs between agents are the quality gates.",[10,22,24],{"id":23},"pm-agent","PM agent",[15,26,27],{},"The PM agent manages the backlog and coordinates the Feed phase.",[15,29,30],{},[31,32,33],"strong",{},"Primary tasks:",[35,36,37,41,49,52,55],"ul",{},[38,39,40],"li",{},"Receives raw requirements from the Orchestrator",[38,42,43,44,48],{},"Creates draft tickets in the ",[45,46,47],"code",{},"Inbox"," state and frames them: title, user story, acceptance criteria, priority, dependencies",[38,50,51],{},"For larger or cross-cutting work, consults the Analytic to co-author a richer task description (still high-level - no technical spec yet)",[38,53,54],{},"Manages dependencies between tickets",[38,56,57],{},"Surfaces prioritisation questions to the Orchestrator when the backlog has conflicting priorities",[15,59,60],{},"The PM agent does not write code, design interfaces, or make technical decisions. Its domain is the shape and flow of the backlog.",[10,62,64],{"id":63},"analytic-agent","Analytic agent",[15,66,67],{},"The Analytic agent is the translator between human intent and technical precision. It plays two distinct roles depending on the phase:",[15,69,70,73],{},[31,71,72],{},"In Feed (advisory, optional):"," For complex tickets, the PM may consult the Analytic to help shape the task description - which areas of the product are affected, which existing use cases interact with the change, what the end-to-end user-visible behaviour looks like. No code-level decisions happen here.",[15,75,76,79],{},[31,77,78],{},"In Forge (mandatory, first step):"," Once the ticket is triggered, the Analytic runs first in the autonomous loop and produces full use-case specifications for every part of the product the change touches. A spec contains:",[35,81,82,85,88,91,94,97],{},[38,83,84],{},"A short narrative of the use case and its actor",[38,86,87],{},"A Mermaid sequence diagram covering the happy path and alternative branches",[38,89,90],{},"The API contract (endpoints, request\u002Fresponse shapes, HTTP status codes)",[38,92,93],{},"Frontend and backend validation tables (field, constraints, size, pattern, notes)",[38,95,96],{},"Data model changes (new entities, columns, migrations)",[38,98,99],{},"A test-case matrix (positive, negative, edge cases)",[15,101,102],{},"The Forge specs are deterministic blueprints that Dev, Reviewer, and Tester all execute against.",[15,104,105],{},"The quality of the Analytic agent's output directly determines the quality of the entire pipeline. A vague spec produces interrupted Forges and rejected Validation Packages. Investing in the Analytic agent's instructions is the highest-leverage activity in the system.",[10,107,109],{"id":108},"dev-agents-fe-be","Dev agents (FE + BE)",[15,111,112],{},"The Dev agents implement the use-case specifications produced by the Analytic in the Forge phase.",[15,114,115],{},[31,116,33],{},[35,118,119,122,125,128],{},[38,120,121],{},"Read the spec and produce working code",[38,123,124],{},"Write unit tests alongside implementation",[38,126,127],{},"Generate or update API documentation",[38,129,130],{},"Handle database migrations",[15,132,133],{},"FE and BE Devs run in parallel, coordinating on shared contracts established at the start of the Forge phase. Neither Dev agent makes architectural decisions that are out of scope for the current ticket - they escalate those via the interruption protocol.",[10,135,137],{"id":136},"reviewer-agent","Reviewer agent",[15,139,140],{},"The Reviewer is the quality police of the Forge loop. It operates after Dev and before DevOps.",[15,142,143],{},[31,144,33],{},[35,146,147,150,153,156],{},[38,148,149],{},"Audits every diff against the technical specification",[38,151,152],{},"Checks for security vulnerabilities, performance regressions, and maintainability issues",[38,154,155],{},"Returns structured failure reports to the Dev when changes are required",[38,157,158],{},"Approves diffs that meet the bar - this approval is the gate to the DevOps phase",[15,160,161],{},"The Reviewer does not implement fixes. It identifies problems with precision and returns them. Dev agents fix and resubmit.",[10,163,165],{"id":164},"tester-agent","Tester agent",[15,167,168],{},"The Tester validates the implementation against the ephemeral environment provisioned by DevOps.",[15,170,171],{},[31,172,33],{},[35,174,175,178,181,184],{},[38,176,177],{},"Executes the full test suite (unit, integration, E2E)",[38,179,180],{},"Reports coverage and any regressions against the baseline",[38,182,183],{},"Returns structured failure reports to the Dev when tests fail",[38,185,186],{},"Approves the suite when all acceptance-criteria tests pass",[15,188,189],{},"The Tester does not write tests for the current ticket from scratch - that is the Dev's responsibility. The Tester executes what exists and validates that nothing has regressed.",[10,191,193],{"id":192},"devops-agent","DevOps agent",[15,195,196],{},"The DevOps agent manages infrastructure, environments, and deployment.",[15,198,199],{},[31,200,33],{},[35,202,203,206,209,212,215],{},[38,204,205],{},"Provisions the ephemeral environment for each ticket in the Forge phase",[38,207,208],{},"Tears down environments after the Validation Package is assembled",[38,210,211],{},"Executes the deployment pipeline on Gatekeeper confirmation",[38,213,214],{},"Monitors the post-deploy health check and reports the result",[38,216,217],{},"Assembles the Validation Package",[15,219,220],{},"The DevOps agent is the last agent to act before the human, and the first agent to act after confirmation. It is the bridge between the autonomous loop and the real world.",{"title":222,"searchDepth":223,"depth":223,"links":224},"",2,[225,226,227,228,229,230,231],{"id":12,"depth":223,"text":13},{"id":23,"depth":223,"text":24},{"id":63,"depth":223,"text":64},{"id":108,"depth":223,"text":109},{"id":136,"depth":223,"text":137},{"id":164,"depth":223,"text":165},{"id":192,"depth":223,"text":193},"The six AI agents that power the Feed, Forge, Finish pipeline and what each one does.","md",{},true,"\u002Fdocs\u002Fai-roles",{"title":5,"description":232},"docs\u002Fai-roles","2kvXvMTziEKCB0mltrfqzBu527wqFjIdSlia8NtX9jY",1778559261707]