Bounded work
Tasks and AgentJobs
The system narrows work into auditable packets with explicit allowed paths, expected outputs, validators, and claim boundaries.
AI research-agent track
AI Research-Agent System
The AI research-agent system is the governed, human-accountable workflow used to plan, check, preserve, and review theoretical physics work. Its workflow evidence can organize research; it does not become physics proof by itself.
System model
Workflow evidence is not physics authority.
The AI system is a research discipline: it makes work inspectable, repeatable, and harder to overstate. Scientific claims still require source artifacts and gates.
Authority
Role contracts
Roles and skills orient work, but task-local execution records and allowlists decide what a specific transaction may do.
Review
Claim gates and refutation
Proposals, stress tests, completions, handoffs, and human gates keep workflow progress separate from physics proof.
Memory
Source-first retrieval
Memory, wiki, registries, and local search help find sources. They do not replace registered source files or authority rows.
Where to go next
Inspect source records before treating a workflow claim as current.
The Phase 5 AI research-agent deep-dive routes are available as a coherent first version after source inspection and QA: workflow, roles and skills, memory and registries, and validator/operator workflow.
Upstream source repositoryCanonical source files, registries, role contracts, and control records remain upstream.Research-agent workflow explainerReviewed generated noncanonical orientation for bounded research workflow.Roles and skills catalogReviewed generated noncanonical guide to role and skill inspection order.
Phase 5A / implemented
Research-agent workflow
/project/ai-research-agent-system/workflow/Workflow explainer for bounded tasks, Director decisions, AgentJobs, completions, and handoffs.Phase 5B / implemented
Roles and skills
/project/ai-research-agent-system/roles-and-skills/Role and skill contract explainer with task-local authority and allowlist boundaries.Phase 5C / implemented
Memory and registries
/project/ai-research-agent-system/memory-registries/Source-first memory, wiki, registry, and retrieval-surface explainer.Phase 5D / implemented
Validator and operator workflow
/project/ai-research-agent-system/validator-operator-workflow/Validator and operator guide that treats PASS results as bounded operational evidence.Source authority
Human accountability remains explicit.
This page can describe the AI-assisted workflow. It cannot expand role authority, change routing behavior, promote physics claims, or replace human-gated publication, authorship, and outreach responsibility.