Journey Day 3

No Man Company Journey

Install to Validation Retrospective — Structuring WSL 2, Docker Sandboxing, NousResearch Hermes Agent, and Hybrid Local LLM Inference

Day 1: Base Platform Setup → Day 2: Core Architecture → Day 3: Agent & Model Integration
57
Total Interactions
14
Milestone Stories
WSL 2
Host Sandbox
Hybrid
Gemini + Ollama
Milestone 1

Host-Level Setup, Roadmaps and Verifications

"Tried to utilize roadmap created prior days, results of YouTube research and conversation with Gemini. Verifying installation by milestones."
#1 open roadmap
Navigation
Exact Prompt:open roadmap
Description:Requested to view the master strategic roadmap.
Agent Action:Listed the workspace files and loaded the contents of roadmap.md.
#2 can I open it as plan right side
UI Config
Exact Prompt:can I open it as plan right side
Description:Requested the roadmap to be rendered as an interactive implementation plan on the side.
Agent Action:Created the implementation_plan.md artifact to render the roadmap in the right-hand panel.
#3 can you check milestone 1.1 ?
Verification
Exact Prompt:can you check milestone 1.1 ?
Description:Requested verification of WSL 2 installation and filesystem path mappings.
Agent Action:Ran status command wsl -l -v and verified that paths map cleanly between host and WSL. Checked off Milestone 1.1 in the plan.
#4 now lets check mailestone 1.2
Verification
Exact Prompt:now lets check mailestone 1.2
Description:Checked the status of Docker Desktop inside WSL.
Agent Action:Ran status checks, identified that Docker was missing/unintegrated, and provided step-by-step guidance.
Milestone 2

Installing & Configuring Docker Desktop

"Verifying architecture and reloading system configuration to enable clean WSL 2 Docker integration."
#5 AMD ARM ?
System Info
Exact Prompt:AMD ARM ?
Description:Clarified system processor architecture for downloading Docker Desktop.
Agent Action:Executed architecture query in PowerShell (resulting in AMD64) and guided user to select the AMD64 download link.
#6 install done, let's go next
Workflow
Exact Prompt:install done, let's go next
Description:Confirmed Docker install and requested next steps.
Agent Action:Checked command paths and guided user to toggle the "WSL integration" settings in Docker Desktop.
#7 I think I restarted, please check now
Verification
Exact Prompt:I think I restarted, please check now
Description:Prompted re-verification of the WSL-Docker integration after settings change.
Agent Action:Ran wsl docker --version, verified integration is functional, and updated the plan. Checked off Milestone 1.2.
Milestone 3

Development Environment & Tool Integration

"Applying strict coding rules and preparing the standard MCP bridges to define the sandbox bounds."
#8 yes please
Confirmation
Exact Prompt:yes please
Description:Approved moving to Phase 2.
Agent Action:Examined and updated CLAUDE.md to enforce strict diff-surgical editing rules.
#9 Option A
Decision
Exact Prompt:Option A
Description:Chosen standard filesystem MCP schemas over a custom wrapper script.
Agent Action:Initialized a standard @modelcontextprotocol/server-filesystem server project under .swarm/mcp-filesystem/.
#10 did we do MS2.2 Wrap all corporate keys, tokens...
Security
Exact Prompt:did we do MS2.2 Wrap all corporate keys, tokens, and social publishing handles inside Google Cloud Secret Manager boundaries.
Description:Checked if sensitive credentials were fully segregated.
Agent Action:Audited the codebase, created a root .gitignore to block credentials, and wrote a Secret Manager integration guide.
Milestone 4

Sandbox and Execution Environment Setup

"Roadmap never had detail on Hermes itself. Demarcating host vs sandbox workspace structures."
#11 proceed to Phase 3
Workflow
Exact Prompt:proceed to Phase 3
Description:Approved moving to Phase 3.
Agent Action:Initialized the file-based transaction folders and the active progress tracking ledger (progress.md).
#12 yes go ahead
Confirmation
Exact Prompt:yes go ahead
Description:Approved creating developer mailbox and QA test runner setup.
Agent Action:Wrote test_runner.js to run QA commands and automatically capture test failures.
#13 did we do 4.1 ?
Security
Exact Prompt:did we do 4.1 ?
Description:Checked if read-only boundaries (Milestone 4.1) were enforced.
Agent Action:Confirmed read-only boundaries are enforced by requiring manual approval before writing files.
#14 Oh my, so this is upper area ,
Architecture
Exact Prompt:Oh my, so this is upper area ,
Description:Identified workspace division between parent and sandboxed environments.
Agent Action:Explained that the orchestrator runs on Windows, while the sandbox acts inside WSL.
#15 Oh my, so this is upper area deploly approved application...
Architecture
Exact Prompt:Oh my, so this is upper area deploly approved application ? So setting up Hermess inside linux is whole seperated story?
Description:Clarified relationship between parent agent and sandboxed Linux agent daemon.
Agent Action:Confirmed setup of Hermes inside Linux/WSL is isolated for secure execution.
Milestone 5

Hermes and Linux Agent Setup

"New plan for Hermes install, node command ran. Managing root privileges inside WSL."
#16 Yes , want to see plan
Confirmation
Exact Prompt:Yes , want to see plan
Description:Requested plan for Linux Agent Sandbox setup.
Agent Action:Created implementation plan for Phase 5 (Linux Agent Sandbox).
#17 answered, lets begin
Workflow
Exact Prompt:answered, lets begin
Description:Approved implementation plan for Phase 5.
Agent Action:Initialized task.md checklist and created wsl_setup.sh inside WSL.
#18 blocked by sudo command or password ?
Trouble
Exact Prompt:blocked by sudo command or password ?
Description:Inquired if the background script was stalling on an interactive password prompt.
Agent Action:Modified the shell script to run without sudo and suggested running it manually as root.
#19 I had no issue , sh
Trouble
Exact Prompt:I had no issue , sh
Description:Confirmed script execution.
Agent Action:Attempted to verify Node version inside WSL.
#20 Should I run wsl_setup.sh myself?
Trouble
Exact Prompt:Should I run wsl_setup.sh myself?
Description:Asked if manual execution in their own terminal was required to authenticate permissions.
Agent Action:Confirmed they should execute the script as root in their own terminal.
#21 Seems all done
Verification
Exact Prompt:Seems all done
Description:Confirmed manual setup commands for Node.js and Git completed successfully.
Agent Action:Checked Node version in WSL, confirmed v22.22.1, and checked off the task.
Milestone 6

Testing and Quality Assurance

"Testing daemon, debugging workspace dependency resolution, and resolving git branch conflicts."
#22 what's next ?
Workflow
Exact Prompt:what's next ?
Description:Asked for the next action to test the agent sandbox.
Agent Action:Developed the agent_daemon.js script and created task_demo_1.json in the developer inbox queue to test polling.
#23 so I can use hermes ?
Clarification
Exact Prompt:so I can use hermes ?
Description:Inquired if Hermes Agent was now usable.
Agent Action:Verified daemon execution which successfully switched branches and processed the mock task.
#24 Command execution failed... enoent Could not read package.json
Trouble
Exact Prompt:Command execution failed... enoent Could not read package.json
Description:Reported file missing error during daemon test.
Agent Action:Explained how to create package.json in the root of the workspace.
#25 some git command is wrong ? ... fatal: a branch named 'dev' already exists
Trouble
Exact Prompt:some git command is wrong ? ... fatal: a branch named 'dev' already exists
Description:Reported branch error in git shell during daemon run.
Agent Action:Corrected git branch logic in agent_daemon.js to handle branch checking safely.
Milestone 7

Validating installation of Hermes

"while validating, it looks different from what I saw in Youtube videos. Inspecting files and codebase configuration."
#26 where is hermess ?
Clarification
Exact Prompt:where is hermess ?
Description:Asked where the Hermes installation lives.
Agent Action:Located the python executable at ~/.local/bin/hermes in WSL.
#27 where is skill and soul ?
Architecture
Exact Prompt:where is skill and soul ?
Description:Asked about modular logic files (skills/soul) of the Hermes configuration.
Agent Action:Located the config directories (~/.hermes/skills/ and system soul templates).
#28 did we use https://github.com/NousResearch/hermes-agent.git
Verification
Exact Prompt:did we use https://github.com/NousResearch/hermes-agent.git
Description:Checked if the official NousResearch Hermes repository was installed.
Agent Action:Verified and confirmed installation points to the official NousResearch hermes-agent v0.16.0 codebase.
Milestone 8

Official Repository Realization & Re-install

"Realized official NousResearch Hermes repository was installed. Never installed, nor planned to install. Starting all over again."
#29 Yes,
Confirmation
Exact Prompt:Yes,
Description:Approved setup strategy for local Ollama backend alongside Gemini.
Agent Action:Created implementation plan for installing Ollama and loading local model worker targets.
#30 yes, let me know step by step instruction
Workflow
Exact Prompt:yes, let me know step by step instruction
Description:Requested setup steps for configuring Hermes API credentials.
Agent Action:Outlined CLI commands to set the paid API key and default planner configurations.
#31 turns out Claude already used up. gemini yet valid...
Key Manage
Exact Prompt:turns out Claude already used up. gemini yet valid, set up done.
Description:Indicated Claude key had hit limit, switching to Gemini instead.
Agent Action:Saved Gemini key and updated the plan to use Gemini 3.5 Flash as primary orchestrator.
Milestone 9

Context Window Specifications & Documentation Feeding

"Fed docs prepared prior. This doc has recommended models per worker. Aligning context limits."
#32 check following docs...
Research
Exact Prompt:check following https://docs.google.com/document/d/1Y9GO3CcG_H15zWNXG3TQwyDfSvnksuWOwQ7ov5Zy0wc/edit?usp=sharing
Description:Provided link to setup references and instructions.
Agent Action:Read the document to extract minimum context length configurations (64k minimum token size).
#33 try again ...
Research
Exact Prompt:try again ...
Description:Resent the document link to ensure successful load.
Agent Action:Read the doc, confirmed requirements (64k minimum token window size limit).
#34 run these agent workers locally on your own hardware...
Architecture
Exact Prompt:run these agent workers (like Qwen/Gemma) locally on your own hardware (e.g., via Ollama in WSL)
Description:Directed setup of local inference engines under WSL.
Agent Action:Structured implementation plan to install Ollama and pull coder models.
#35 updated comments
Verification
Exact Prompt:updated comments
Description:Selected setup parameters and approved trial of 14B+ models.
Agent Action:Created wsl_install_ollama.sh script to configure systemd and start daemon.
#36 it's done,
Verification
Exact Prompt:it's done,
Description:Confirmed Ollama installation complete.
Agent Action:Tested localhost endpoint http://localhost:11434 to verify service is running.
Milestone 10

VRAM/RAM Exhaustion & Tuning Model Tags

"Installing or pulling models. Somehow Gemini pulled older models ignoring document. Switching to latest tags."
#37 why you are pulling gemma 2 instead 4 and qwen2.5 instead 3.6 ?
Model Strategy
Exact Prompt:why you are pulling gemma 2 instead 4 and qwen2.5 instead 3.6 ?
Description:Challenged older local model versions.
Agent Action:Looked up latest models in Ollama Library and updated commands to pull qwen3.6:35b and gemma4:31b.
#38 restart done,
Verification
Exact Prompt:restart done,
Description:Confirmed WSL reboot completed.
Agent Action:Checked off systemd and driver tasks in the list.
#39 will gemma4 also work smaller one
Model Strategy
Exact Prompt:will gemma4 also work smaller one
Description:Asked if smaller model parameter sizes could be used to prevent crash.
Agent Action:Recommended smaller parameter versions (e.g. e4b) to save VRAM.
#40 plan was to have different models... how is setup now?
Architecture
Exact Prompt:plan was to have different models for worker agents. how is setup now?
Description:Inquired about worker vs primary orchestrator model routing.
Agent Action:Explained model delegation inside config.yaml (Gemini as planner, Ollama for workers).
#41 is gemini-1.5-pro correct for gemma 4?
Architecture
Exact Prompt:is gemini-1.5-pro correct for gemma 4?
Description:Asked if Gemini 1.5 Pro should route tasks to Gemma 4.
Agent Action:Clarified Gemini 1.5 Pro is the cloud brain that spawns subagents who talk to local Gemma 4.
#42 i've set gemini_api_key, let's use best available.
Key Manage
Exact Prompt:i've set gemini_api_key, let's use best available.
Description:Stated API key is set and ready to use.
Agent Action:Configured orchestrator backend.
Milestone 11

Gemini 3.5 Flash Orchestrator & Paid API Quota

"I had paying plan for Gemini so asked to use Gemini-3.5-Flash for upper agent."
#43 want to use gemini-3.5-flash)
Model Strategy
Exact Prompt:want to use gemini-3.5-flash)
Description:Indicated preference for Gemini 3.5 Flash.
Agent Action:Configured model.default to gemini-3.5-flash.
#44 can we test ?
Verification
Exact Prompt:can we test ?
Description:Triggered model inference test.
Agent Action:Executed test command, encountered HTTP 429 quota limits.
Milestone 12

Local Worker Delegation & Secret Manager Key Alignment

"Asked to setup lower agents to use gemma4. Had to go Google AI Studio and more to fix API key."
#45 what about gemma4 for worker ?
Model Strategy
Exact Prompt:what about gemma4 for worker ?
Description:Checked setup of local Gemma 4 worker model.
Agent Action:Configured local worker parameters in config.yaml.
#46 limits to 5 per minute... can you check below ?
Trouble
Exact Prompt:I have given API Key for Gemini can you check below ? On the free tier, Google limits Gemini 3.5 Flash requests to 5 per minute.
Description:Shared research about Gemini free tier limitations causing the HTTP 429 crash.
Agent Action:Suggested switching to 1.5 Flash or upgrade.
#47 I am saying I have subscription
Trouble
Exact Prompt:I am saying I have subscription
Description:Confirmed active paid billing subscription.
Agent Action:Explained how to generate API key under billed project in AI Studio.
#48 fixed, want to double check where is key stored?
Security
Exact Prompt:fixed, want to double check where is key stored?
Description:Confirmed key update and asked for storage path.
Agent Action:Pointed to ~/.hermes/.env where credentials are loaded.
Milestone 13

Hybrid Verification & QA Loop execution

"now testing all again DEV-QA loop."
#49 now check
Verification
Exact Prompt:now check
Description:Re-tested model verification after key upgrade.
Agent Action:Confirmed successful hybrid execution (Gemini 3.5 Flash orchestrator + local Gemma 4 worker).
Milestone 14

Dashboard Setup, Symlink Overrides & PATH Isolation

"Now asking continuous loop and dashboard. Had to go through troubleshooting to fix dashboard. It continued to next day."
#50 Great, I just want to test continuesly run project...
Feature Test
Exact Prompt:Great, I just want to test continuesly run project, and watch hermes doing provisioing, and Dev - QA back and forth. any idea ?
Description:Requested loop demonstration run.
Agent Action:Wrote run_dev_qa_demo.sh to trigger Junior Dev / QA error loop simulation.
#51 tell me more on (Meanwhile, you can keep the hermes dashboard...)
UI/UX
Exact Prompt:tell me more on (Meanwhile, you can keep the hermes dashboard running in a separate tab to watch the active processes in your browser!)
Description:Requested dashboard features overview.
Agent Action:Explained active tree logs, live logs, task board, and MCP schemas.
#52 hermes dashboard building failed... call stack size exceeded
Trouble
Exact Prompt:hongw@DESKTOP-984B3DB:~/.hermes/hermes-agent/venv/bin$ hermes dashboard ... npm install failed ... call stack size exceeded
Description:Reported npm build issues with WSL symlink bounds.
Agent Action:Diagnosed Windows-native npm pollution and guided user to use native Linux npm.
#53 npm install --workspace web ... EISDIR: illegal operation
Trouble
Exact Prompt:npm install --workspace web && npm run build -w web ... EISDIR: illegal operation on a directory, symlink
Description:Reported symlink failures when building UI manually.
Agent Action:Explained how to run npm with absolute Linux path /usr/bin/npm to bypass Windows path overrides.
#54 dashboard is working , what was next ?
Workflow
Exact Prompt:dashboard is working , what was next ?
Description:Confirmed dashboard successfully launched.
Agent Action:Guided user to open browser and run loop scripts.
#55 half way
Trouble
Exact Prompt:half way
Description:Dashboard background builder still failing due to Windows path resolution.
Agent Action:Suggested stripping Windows entries (/mnt/c/) from execution PATH to resolve it once and for all.
#56 give me my prompt asking dashboard and dev qa loop
Info Retrieve
Exact Prompt:give me my prompt asking dashboard and dev qa loop
Description:Requested input text for visual dashboard run.
Agent Action:Provided a code test prompt with division-by-zero bug instructions.
#57 na, what was my ask exactly ? just copy as id
Info Retrieve
Exact Prompt:na, what was my ask exactly ? just copy as id
Description:Requested exact prompt text.
Agent Action:Copied their exact prompt text from transcript logs.