Less one-off prompting, more reusable operating system for agent work.
Agent workflow timeline
From editor helper to routed agent cockpit.
A visual timeline of how the workflow moved from inline Copilot help to CLI agents, vault-backed context, Hermes routing, Codex coding loops, and reusable skills.
4 stages2025 → 2026repo + vault contextskills as memory
Repos and vault context beat chat memory. The public pages explain the durable layers.
Work is not done until it is run, visually checked where needed, committed, and reported with evidence.
Each stage is a public explanation of a workflow layer, not a full changelog. The pages name the shift, the useful output, and the limitation that pushed the next stage.
Agent Workflow is a system evolution page. The reusable rule for running agent work lives under Methodologies as Agent Workflow Loop; individual procedures live under Skills.
Public copy stays high-level: no private vault contents, no fake metrics, and no claim that a stage was perfect just because it existed.
Evolution
2025 December
GitHub Copilot in VS Code
AI starts as an editor-side acceleration layer.
Existing editor workflowFaster snippets
2026 January
Claude Code CLI & Agent Brain
The workflow becomes command-line and task-shaped.
Repo filesReusable command patterns
2026 February
Claude Code & life-os-x
Agent work connects to a markdown-first vault system.
Vault notesA vault-backed way to preserve context
2026 May
Hermes, Codex & viggo-agent-skills
The stack becomes routable, reusable, and multi-agent.
Telegram commandsA practical multi-agent operating model with reusable skill memory