New Skills: June 22, 2026 (Late Sweep) -- aawobdev/hermes-skills¶

Discovered: June 22, 2026 via GitHub search + skills.sh New repo: 1 | Net-new skills: 12 Repo: aawobdev/hermes-skills

A complete multi-agent blueprint orchestration system that splits work across 9 specialized roles -- an expensive thinking model produces a detailed blueprint, then cheap local models execute under human supervision.

⭐ aawobdev/hermes-skills -- Blueprint Orchestration¶

Stars: 0 ⭐ | License: Not specified | Created: May 27, 2026 Author: Alistair (aawobdev) Languages: Python 86.9%, Shell 13.1% Commits: 83 | Status: Actively maintained (latest commit: June 22, 2026)

Core Philosophy¶

"The expensive model thinks. The cheap models do. You supervise."

A frontier thinking model acts as Architect -- it interviews you, makes all design decisions, and produces a detailed blueprint. Cheap local models then execute the blueprint role-by-role. Roles never talk directly -- you are the relay.

"If the blueprint is thorough enough, you don't need a frontier model for every task. A blueprint that specifies exact file names, function signatures, and verification commands turns a $0.001/task local model into a reliable executor."

Skills Included (12)¶

#	Skill	Role	Description
1	`blueprint-orchestration`	Entry Point	Full workflow: interview → blueprint → one-shot execution. Start here.
2	`role-architect`	Architect	System prompt: interviews, designs, produces blueprints, patches specs
3	`role-designer`	Designer	System prompt: visual decisions, UI layouts, interaction patterns
4	`role-developer`	Developer	System prompt: executes build tasks, writes code, never decides
5	`role-tester`	Tester	System prompt: adversarially verifies output against the spec
6	`role-devops`	DevOps	System prompt: CI/CD, environments, release, rollback, observability
7	`role-researcher`	Researcher	System prompt: gathers evidence, compares options, produces reports
8	`role-security-auditor`	Security	System prompt: vulnerability review, credentials, attack surface
9	`role-end-user`	End-User	System prompt: simulates real user to find UX gaps
10	`role-orchestrator`	Orchestrator	System prompt: sequences tasks, delegates via one-shot commands
11	`prompting-standards`	Standards	LLM/prompt best practices for authoring + executing blueprints
12	`model-routing`	Routing	Model roster, VRAM constraints, role-to-model assignments (Ollama + OpenRouter + Claude Code)

The Workflow¶

You → Architect (thinking model) → Blueprint
         ↑                              ↓
    Researcher                 ┌─────────────────────────┐
  (evidence reports)           │ Developer (cheap model)  │ ← executes tasks
                               │ Tester (mid model)       │ ← validates output
                               │ Security (thinking model)│ ← reviews for risks
                               │ End-User (any model)     │ ← simulates usage
                               └─────────────────────────┘
                                           ↓
                                   You sign off

Phases: 0: Interview → 1: Blueprint → 2: Design → 3: Build → 4: Test → 5: Deploy → 6: Security Audit → 7: End-User Review → 8: Closeout

Any escalation goes back to Architect. Human is the relay between all phases.

Why This Matters for Hermes Operators¶

This is the first community blueprint orchestration system we've discovered for Hermes Agent. It addresses a core pain point: how to structure multi-agent projects without a single omnipotent prompt. Key innovations:

Model routing by tier -- TIER 0 (OpenRouter Orchestrator) → TIER 1 (local Ollama for routine dev) → TIER 2 (OpenRouter paid for complex tasks) → TIER 3 (OpenRouter free tier) → TIER 4 (Claude Code CLI). Role-appropriate model assignment based on task complexity.
Self-contained blueprints -- Each role receives only its task section + role card. Zero context leaking between roles.
STATUS.md tracking -- Every task starts as ⬜ Todo, progresses through 🟦 In Progress → ✅ Done → ❌ Failed. Live progress, not a static plan.
Escalation protocol -- Roles escalate to Architect (via human relay) when stuck, confused, or the spec is ambiguous. No silent drift.
Local-first, cloud-fallback -- Uses local Ollama models by default (free, private), falls back to OpenRouter only for reasoning-heavy tasks.

Installation¶

Option A -- Hermes Tap (recommended)¶

hermes skills tap add aawobdev/hermes-skills

Skills are immediately available. Update anytime:

hermes skills tap update aawobdev/hermes-skills

Option B -- Manual Clone¶

git clone https://github.com/aawobdev/hermes-skills ~/hermes-skills

Then in ~/.hermes/config.yaml:

skills:
  external_dirs:
    - ~/hermes-skills/skills

To Start a Project¶

/skill blueprint-orchestration

My project: I want to [one sentence description]

Model Defaults¶

Role	Primary Model	Runtime
Architect	`qwen3.6-35b-a3b`	LM Studio
Designer	`qwen3.6-35b-a3b` (UX reasoning) + `gemma4:26b` (vision) + Claude Sonnet (creative)	Mixed
Developer	`qwen3-coder-30b`	LM Studio
Tester	`gemma4:26b`	Ollama
DevOps	`qwen3-coder-30b`	LM Studio
Researcher	`qwen3.6-35b-a3b` or Claude Sonnet	LM Studio / Claude
Security	`qwen3.6-35b-a3b`	LM Studio
End-User	`gemma4:26b`	Ollama
Orchestrator	`deepseek/deepseek-v4-flash`	OpenRouter

Full model roster (24 models across Ollama + OpenRouter + Claude Code CLI) is documented in model-routing.

Summary¶

Metric	Value
New repos	1
New skills	12
Skill categories	Orchestration, Roles, Standards, Routing
First-of-kind	Multi-agent blueprint orchestration for Hermes
Setup guide	blueprint-orchestration-setup.md

← Previous Discovery | Marketplace Home →