Hermes Precision Pack

I just turned a massive Claude-style system prompt into a clean, modular Hermes setup. Introducing the Hermes Precision Pack. Built for people who want Hermes to feel less like a generic chatbot and more like a focused AI operator that actually understands how they work.

Inside the pack:

• A custom SOUL.md for personality and behavior
• USER.md and MEMORY.md for persistent context
• A reusable AGENTS.md project template
• Modular Hermes skills
• Installation and verification scripts
• Backup-safe setup
• Migration notes and documentation

I also stripped out everything Claude-specific, including incompatible tool schemas, Anthropic identity rules, Artifact APIs, and broken runtime paths. The result is leaner, cleaner, and ready for real-world use.

No giant mega-prompt. No unnecessary prompt bloat. No "installed successfully" while quietly doing absolutely nothing. Just a proper Hermes foundation you can build on.

I'll be testing and expanding this into a more complete local AI memory, workflow, and agent system next. Hermes is about to get dangerously useful.

What it does

The Precision Pack gives Hermes a consistent identity, a sharp communication style, well-structured memory, and a battle-tested workflow for complex multi-step tasks.

Without it, Hermes runs generic. With it, Hermes operates with Ryan's context loaded — workspace habits, PPC stack conventions, preferred communication style, and a precision workflow that catches errors before they compound.

Tech Stack

Component	Technology
Skill Format	Hermes Markdown (Markdown)
Config	YAML
Templates	SOUL.md, USER.md, MEMORY.md, AGENTS.md
Scripts	Bash
Compatibility	Hermes Agent (Nous Research)

📊 Codegraph

See codegraphs/hermes-precision-pack.md for the full dependency graph.

---

What's inside

SOUL.md — Identity and behavior

Defines Hermes' core posture: pragmatic, direct, warm but not performative. Sets the communication contract — concise answers, no filler, admits uncertainty, pushes back when something is wrong. This is the foundational layer that shapes every response.

memories/USER.md — Ryan's profile

Compact user profile within Hermes' character limits. Covers communication style preferences, work habits, and what Ryan values in deliverables. Loaded on every session so Hermes never starts cold.

memories/MEMORY.md — Operational context

Durable facts about Ryan's environment: PPC stack (ClickUp, MerchantSpring, Adbrew, Helium 10, Data Dive), workstation specs, timezone, and domain-specific conventions like branded term handling, campaign structure patterns, and spend guardrails. Updated proactively as new patterns emerge.

skills/precision-workflow/ — Execution workflow

Reusable multi-step workflow with five reference documents:

Reference	Use when
research-and-verification.md	Looking up facts, current data, or anything requiring sources
file-and-artifact-work.md	Creating, editing, or validating files and scripts
long-task-execution.md	Multi-file changes, code audits, extended automation work
memory-and-learning.md	Deciding what to remember vs. discard after a task
safety-and-boundaries.md	Credentials, personal data, high-stakes topics

Trigger the workflow explicitly with /precision after installing the skill bundle.

skill-bundles/precision.yaml — /precision shortcut

Bundles the precision workflow under a single command. After installation, type /precision <task> to activate the full workflow.

project-template/AGENTS.md — Project operating rules

Drop this into any project root to give Hermes context-aware execution rules: inspect-before-changing, minimal complete changes, verifiable outputs, and documentation discipline. Subdirectory AGENTS.md files take priority, so project-specific rules override the template.

config.precision.snippet.yaml — Config overlay

Four keys to merge into ~/.hermes/config.yaml:

memory:
  memory_enabled: true
  user_profile_enabled: true
  memory_char_limit: 2200
  user_char_limit: 1375
  write_approval: true   # prompts before memory writes

skills:
  write_approval: true    # prompts before skill installs

display:
  interim_assistant_messages: true
  file_mutation_verifier: true   # prompts before file changes
  memory_notifications: verbose
  compact: false

The write_approval gates are the most important additions — they prevent runaway memory inflation and catch accidental skill bloat. Leave them on until you've established clear trust in the workflow.

---

Quick install

# 1. Extract the pack
unzip hermes-precision-pack-*.zip -d hermes-precision-pack
cd hermes-precision-pack

# 2. Install (backups existing files first)
chmod +x install.sh verify.sh
./install.sh

# 3. Verify
./verify.sh

# 4. Merge the config snippet (optional but recommended)
# Edit ~/.hermes/config.yaml and add the keys from global/config.precision.snippet.yaml

# 5. Restart your Hermes session

With a specific Hermes home

./install.sh --home /path/to/hermes/home

With project-level rules

./install.sh --project /path/to/your/project

This copies project-template/AGENTS.md into the project root.

---

File map

hermes-precision-pack/
├── README.md              ← you are here
├── MIGRATION-NOTES.md     ← what changed and why
├── MANIFEST.sha256        ← integrity checksums
├── install.sh             ← backup-aware installer
├── verify.sh              ← post-install checks
├── global/
│   ├── SOUL.md            ← identity and communication rules
│   ├── config.precision.snippet.yaml  ← config keys to merge
│   ├── memories/
│   │   ├── USER.md        ← Ryan's profile (compact)
│   │   └── MEMORY.md      ← operational context (compact)
│   ├── skill-bundles/
│   │   └── precision.yaml ← /precision shortcut bundle
│   └── skills/
│       └── precision-workflow/
│           ├── SKILL.md   ← workflow with task routing
│           ├── agents/
│           │   └── openai.yaml
│           └── references/
│               ├── research-and-verification.md
│               ├── file-and-artifact-work.md
│               ├── long-task-execution.md
│               ├── memory-and-learning.md
│               └── safety-and-boundaries.md
└── project-template/
    └── AGENTS.md          ← drop into project roots

---

Integrity verification

Every file in the pack is checksummed in MANIFEST.sha256. Verify after installation:

cd hermes-precision-pack
sha256sum --check MANIFEST.sha256

All files should report OK. Any FAILED line means the file was modified or corrupted — do not use it and reinstall from a clean extraction.

---

Uninstall / restore

The installer backs up every file it replaces into:

~/.hermes/backups/precision-pack-<timestamp>/

To restore a specific file:

cp ~/.hermes/backups/precision-pack-<timestamp>/<path> ~/.hermes/<path>

To fully uninstall, delete the installed files:

rm ~/.hermes/SOUL.md
rm ~/.hermes/memories/USER.md
rm ~/.hermes/memories/MEMORY.md
rm -rf ~/.hermes/skills/precision-workflow
rm ~/.hermes/skill-bundles/precision.yaml
rm ~/.hermes/config.precision.snippet.yaml

---

Updating the pack

When a new version is released:

1. Download and extract the new pack.
2. Run ./install.sh — it backs up the current version automatically.
3. Run ./verify.sh to confirm.
4. Restart your Hermes session.
5. Check MIGRATION-NOTES.md for any structural changes.

---

Recommended config merge

The installer does not touch your config.yaml. To get the full benefit, add these keys from global/config.precision.snippet.yaml:

# ~/.hermes/config.yaml

memory:
  # existing keys...
  memory_enabled: true
  user_profile_enabled: true
  memory_char_limit: 2200
  user_char_limit: 1375
  write_approval: true

skills:
  # existing keys...
  write_approval: true

display:
  # existing keys...
  interim_assistant_messages: true
  file_mutation_verifier: true
  memory_notifications: verbose
  compact: false

---

Using the precision workflow

After installation, invoke the full workflow on any complex task:

/precision audit this repository and fix the highest-risk issues

Or let Hermes load it naturally when the task matches — long research, multi-file changes, file creation, audits.

---

Compatibility

• Hermes — designed for Hermes v0.9+ with skills, memory, and holographic memory subsystems.
• Platforms — tested on Linux (VPS), Android/Termux. Portable shell scripts, no desktop-only dependencies.
• Character limits — USER.md is 976/1375 chars. MEMORY.md is 1337/2200 chars. Both pass the verification script.

---

What's in USER.md vs MEMORY.md

	USER.md	MEMORY.md
Purpose	Who Ryan is, how he communicates	What's true about Ryan's environment
Contents	Name, timezone, preferences, coding style	Workstation, PPC stack, conventions, guardrails
Changes when	Ryan tells you a new preference	You discover a new fact about the setup
Format	First-person style, directly readable	Compact facts, §-separated

---

Credits

Built from behavioral analysis of a source system prompt, stripped of model-specific artifacts, and adapted for the Hermes agent framework by Ryan Dabao.

The workflow draws from research-backed execution patterns: verification before claims, minimal complete changes, progressive disclosure, and explicit quality gates.