← Back to portal

FIELD NOTES // DOSSIER

OPERATOR DOSSIER

Operator dossier: how a QA engineer moved from web chat as “smart search” in 2023 to a multi-agent pipeline. Snapshot numbers, stack, and the next step. Some early sessions were cleaned from logs; those bits are restored from practice memory.

Snapshot 13 July 2026Operator: Denis Sviridov

01 · SIGNALS

Four numbers on the snapshot date — not live telemetry.

  • CURSOR SESSIONS772Cursor agent chats across projects on this machine
  • AGENT FLEET1102Sum of sessions/tasks: Cursor, OpenCode, Hermes, Pi, and more
  • PORTAL COMMITS309Committed work in PersonalBlog
  • STATUSFIELD OPSOperating mode for this snapshot — not a game rank

02 · BRIEFING

Why this page exists and where the snapshot comes from.

BRIEF

2023: ChatGPT in the browser as search amplification (Google/Yandex). Late 2024: DeepSeek + paste into the IDE → a Telegram Mini App Advent calendar for the company QA community. 2025: agents, MCP, RAG. 2025/2026 boundary: company coverage dashboard in Cursor (early skills + custom MCP). 2026: runtime fleet and governed ops; the personal portal is one artifact next to automation and own AI projects. Session counts are a lower bound — part of the history lived only in chats and was cleaned.

03 · OPERATOR PROFILE

What already holds in production — and what still needs you in the chair.

Strengths

  • Builds pipelines across agents and MCP — not one endless chat
  • Agree on a spec/plan before coding
  • Picks the runtime for the job: Cursor, OpenCode, Hermes, Pi…
  • Ships to prod: commits, deploy, live product
  • Doesn't take the agent at its word — review, tests, security/SEO

Growth areas

  • Public cases and talks so others can copy the practice
  • Short “how to repeat” playbooks for a team
  • English dossier and reach beyond RU
  • Live pipeline demos in front of people, not only screenshots

04 · SKILL EVOLUTION

How the way of working with AI changed since 2023 — from chat tips to a governed fleet. About skill, not website features.

  1. 2023

    Chat as smarter search

    2023: ChatGPT in the browser. Not “build a product” — amplify search: better queries, quick checks, fewer Google/Yandex tabs.

  2. late 2024

    First AI-shipped artifact

    Late 2024: DeepSeek in the browser and paste into the IDE. First result that reached people — a Telegram Mini App Advent calendar for the company QA community.

  3. 2025

    Agent lab

    2025: Claude Code, Cursor, VS Code/PyCharm extensions (Kilo Code, Continue, Firebender…). In parallel MCP (Playwright among the early servers) and RAG. A “map the zoo” year, not a forever lock-in to one tool.

  4. late 2025 – early 2026

    Skills + company dashboard

    Late 2025 – early 2026: company coverage dashboard in Cursor with the first skills. TestOps, DocHub, GitLab, SonarQube; per-service cards; admin UI and a custom MCP so an agent could pull stats.

  5. spring–summer 2026

    Runtime fleet

    Spring–summer 2026: OpenCode, Hermes, Pi, Cline, and more in daily work. Cockpit per job — IDE, TUI, desktop, CLI. One operator, different runtimes.

  6. 2026-07

    Governed ops

    Jul 2026: spec → plan → verify, MCP into work systems, Codegraph, review skills, portal as proven ship. The agent is not magic — a pipeline step with a Definition of Done.

05 · WHERE THE WORK HAPPENED

Top Cursor projects by session count, plus key milestones.

  • Hoff tests (automation)

    141 sessions

    Hoff retail automation suite: scenarios, fixtures, failure triage. AI to write/fix tests and move faster through a large suite.

  • PersonalBlog (personal portal)

    125 sessions

    Blog, admin, client portal, CRM — agents from features to deploy. Also the lab for skills, MCP, and agent rules.

  • Hoff app test (mobile)

    124 sessions

    Mobile app test automation: structure, fixtures, flaky failures. Heavy “agent + real test repo” use.

  • QA Agent (own agent)

    103 sessions

    Configurable QA agent: CLI/UI, LLM providers, MCP/tools, flows for cases and checklists. Building agents, not only using them.

  • Coverage dashboard (company)

    milestone

    Coverage dashboard: TestOps + DocHub + GitLab + SonarQube, per-service cards, admin UI, and MCP. Cursor + early skills at the 2025/2026 boundary — bridge from experiments to in-company AI ops.

06 · TOOLING

Two layers: where the conversation ran (runtimes) and where the agent fetched data (MCP / skills).

Agent runtimes

Where the model + tools conversation ran.

  • Cursor
  • OpenCode
  • Hermes
  • Pi
  • Claude Code
  • Codex
  • Cline
  • Kilo Code
  • Continue
  • Firebender

MCP → external systems

The agent hits work systems, not only local files.

  • Playwright MCP
  • Jira
  • Confluence
  • TestOps
  • Grafana
  • GitLab
  • Codegraph
  • AgentMemory

Skills & rules

Reusable instructions: review, tests, writing.

  • Skills (SKILL.md)
  • AGENTS.md rules
  • Security review
  • SEO audit
  • Playwright checks
  • RAG experiments

How work got shipped

Process so AI work doesn't sprawl.

  • Spec → plan → ship
  • TDD / verify before done
  • Multi-agent review
  • Docker deploy

FLEET BREAKDOWN

How many sessions where — and why the tool is on the list.

  • Cursor772

    Main IDE: code, refactors, long features.

  • OpenCode172

    TUI/desktop: portal work, MCP, audits, parallel task lane.

  • Hermes43

    Desktop runtime: digests, experiments, QA-related flows.

  • Kilo Code24

    VS Code / Windsurf agent for focused editor tasks.

  • Pi23

    CLI: worktrees, Hoff/PersonalBlog, fast terminal ops.

  • Cline21

    Extension with task sessions: edits per task.

  • Claude Code19

    Anthropic terminal agent: separate repo sessions.

  • Factory14

    Another runtime in the zoo — comparing UX and pipelines.

  • Codex13

    OpenAI Codex CLI: short/medium coding sessions.

  • Crush1

    Light runtime trial (barely used).

07 · WHAT'S NEXT

Steps toward handoff: so another engineer can repeat the practice.

  1. 2–3 case studies “problem → agents → outcome” with time/quality numbers.
  2. A short playbook: skills + MCP + Definition of Done.
  3. A talk (internal or external) based on this dossier.
  4. Track lead time, defects, coverage — not only session counts.
  5. Coach one engineer through spec → plan → ship.
  6. Refresh this snapshot quarterly with new fleet numbers.

08 · SNAPSHOT TAKEAWAY

Where the practice sits now — and the next ops goal.

FIELD OPERATOR · ACTIVE

Operator in the field: from 2023 (chat as search) → Mini App → agent/MCP/RAG lab → company dashboard with skills → runtime fleet (~1100+ saved sessions as a lower bound) and a live portal. Next step — handoff: the system can be transferred and repeated without you in the operator chair.

NEXT OPS → HANDOFF READY

Session numbers are a local log snapshot on the report date, not live telemetry. 2023–2025 is partly restored by hand: chats were cleaned, and some work lived only in web UIs without agent transcripts. Tools define “session” differently.