Proof by default
Operational proof is the product primitive.
Every workflow run should show what happened, what was approved, what artifacts were produced, what checks passed, what failed, and what should happen next.
Agent Run Pack
Objective in. Controlled run. Proof out.
The first build target after landing validation is a portable run packet that makes agent work reviewable, repeatable, and improvable.
agentfoundry-run/ objective.md runtime-plan.yaml tool-policy.yaml traces/ artifacts/ eval-report.md approval-log.md proof-packet.md next-actions.md
01Objective
One workflow with business outcome, scope, policy, success criteria, and stop conditions.
02Runtime
Agents, tools, MCP connectors, browser/code execution, memory, skills, and human checkpoints.
03Governance
Tool permissions, approval rules, audit trails, rollback plans, risk flags, and policy gates.
04Proof
Traces, evals, artifacts, receipts, handoffs, and a reusable run packet.
05Learning
Failures become evals, policies, prompts, skills, and reusable workflow recipes.
TraceStep-by-step tool calls, execution state, inputs, outputs, timestamps, and replay notes.
ApprovalHuman gates for risky, external, destructive, production, financial, or policy-sensitive actions.
EvalAcceptance checks, regression cases, policy checks, validation reports, and failure explanations.
ArtifactDiffs, reports, cited evidence, drafted responses, datasets, handoff notes, and deployment receipts.
Proof packetRun summary, risks, outcomes, approvals, receipts, next actions, and improvement backlog.