create-company-ai-employees

Create Company AI Employees Skill

Build a reusable skill that generates a company-specific AI employee organization — tailored profiles, SOUL templates, Kanban board, cron jobs, and full Hermes integration — based on company research (tech stack, org structure, products, culture).

What it does: Takes company DeepResearch output and produces a complete hermes skills install-able skill directory with 5–8 specialized profiles, each with a SOUL.md tuned to the company's domain, stack, and processes.

What it does NOT do: Run the agents, provision cloud infra, or replace human engineers. It only authors the skill artifact.

Dependencies: company-deepresearch skill (for research), ai-employee-org (base patterns), skill_manage (authoring). Stdlib + Hermes tools only.

When to Use

• "Create AI employee org for [Company X] based on our research"
• "Generate LeadingMark-style skill for another HR Tech SaaS"
• "Bootstrap autonomous agent team matching a company's tech stack and org chart"
• "Turn DeepResearch brief into runnable Hermes skill"

Prerequisites

• Hermes with skill_manage tool available
• company-deepresearch skill installed (for research phase)
• ai-employee-org skill installed (for base templates)
• Target company DeepResearch brief (Markdown with: tech stack, org structure, products, culture, metrics)
• Write access to ~/.hermes/skills/<category>/<skill-name>/

How to Run

# 1. Research the target company (if not done)
hermes skill company-deepresearch run --company "Target Company"

# 2. Invoke this skill authoring workflow
# Option A: Via execute_code (orchestrated)
hermes execute_code "$(cat scripts/create-company-ai-employees.py)"

# Option B: Step-by-step through chat using skill_manage
# (see Procedure below)

Quick Reference

Step	Tool	Output
Research	company-deepresearch	brief.md
Analyze	execute_code / chat	profile list, stack mapping
Author SKILL.md	skill_manage create	skill skeleton
Write SOUL templates	skill_manage write_file	templates/SOUL-*.md
Verify	skill_view	loadable skill

Procedure

1. Gather Company Intelligence

Run DeepResearch to produce a structured brief:

# Using the company-deepresearch skill pattern
# Output: brief.md with sections:
# - tech_stack: {backend, frontend, infra, ml, data}
# - org_structure: {teams, roles, headcount, leadership}
# - products: [{name, domain, challenges}]
# - culture: {values, rituals, evaluation, autonomy}
# - metrics: {slo, cost_targets, velocity, quality_gates}

2. Map to AI Employee Profiles

Derive 5–8 profiles from org structure + tech stack:

Source Signal	Profile	Example
ML team + model serving	*-ai-engineer	Inference, MLOps, GPU fleet
Research-heavy (ArXiv, experiments)	*-ml-researcher	Survey, experiment design, handoff
Platform/SRE/infra	*-devops-engineer	EKS, Terraform, observability, cost
Product engineering (FE/BE)	*-product-engineer	Rails/React, sprints, feature flags
Security/compliance focus	*-security-engineer	AppSec, DataSec, MLSec, audit
Orchestration/human interface	*-secretary	Triage, decompose, approvals
Continuous improvement	*-self-improver	Curator, memory, config drift

Rule: One profile per distinct accountability domain, not per human role. Merge if same tools/cadence.

3. Author SKILL.md

skill_manage create \
  --name "<company-slug>-ai-employees" \
  --category "autonomous-ai-agents" \
  --content "$(cat <<'EOF'
---
name: <company-slug>-ai-employees
description: "AI employee roles tailored for <Company> <domain> engineering team."
version: 0.1.0
author: Hermes
license: MIT
platforms: [linux, macos, windows]
metadata:
  hermes:
    tags: [<Company>, AI-Employee-Org, Kanban, Autonomous-Agents, <Domain>]
    category: autonomous-ai-agents
    related_skills: [ai-employee-org, hermes-agent, company-deepresearch, openmanus-delegation]
    profiles:
      - <profile-1>
      - <profile-2>
      # ... 5-8 total
EOF
)"

4. Write SOUL Templates (one per profile)

For each profile, create templates/SOUL-<profile>.md with:

# <Profile Title> — <Japanese Role>

You <one-sentence mission>.

## Mission
- Bullet 1: core accountability
- Bullet 2: key deliverable
- Bullet 3: quality standard

## Execution
- **Stack**: exact technologies from brief
- **Tools**: `terminal`, `kanban`, `delegate_task`, `cron`, `web`
- **Cadence**: sprint / cron / event-driven
- **Definition of Done**: measurable gates

## Product-Specific Context (if applicable)
| Product | Key Technical Challenges |
|---------|-------------------------|
| ...     | ...                     |

## Collaboration
- Receives from: <other profiles>
- Sends to: <other profiles>
- Escalates: <conditions>

## Complete Metadata
```json
{
  "changed_files": [],
  "verification": "measurable success criteria",
  "deliverable": "concrete artifact"
}

### 5. Add Procedural Sections to SKILL.md

Append to SKILL.md:

- **How to Run**: bootstrap commands (profile create, skill install, SOUL copy, kanban init, cron install)
- **Quick Reference**: profile table with tools/cadence
- **Procedure**: work enters → completes → periodic autonomy
- **Pitfalls**: delegate_task not durable, WIP limits, one dispatcher
- **Verification**: checklist commands

### 6. Validate

```bash
hermes skill_view <company-slug>-ai-employees
hermes -p <profile> skills list | grep <company-slug>
hermes -p <profile> kanban list --status ready,running

Pitfalls

• Profile explosion: >8 profiles creates coordination overhead. Merge adjacent domains.
• Generic SOULs: Every template must reference company-specific products, stack, metrics. No boilerplate.
• Missing cron integration: Each profile needs at least one cron job for autonomy.
• Skill size: Keep SKILL.md ~200 lines; large scripts → scripts/, references → references/.
• Secret handling: Never bake API keys. Document ANTHROPIC_API_KEY / OPENAI_API_KEY / etc. as prerequisites.

Verification

Skill is valid when:

# All checks pass
hermes skill_view <skill-name>  # loads without error
hermes profile list | grep -c <prefix>  # returns profile count (5-8)
ls ~/.hermes/skills/.../templates/SOUL-*.md | wc -l  # matches profile count
grep -r "company-specific" ~/.hermes/skills/.../templates/  # finds product/stack refs

---

Example Output: LeadingMark AI Employees

The skill we just created (leadingmark-ai-employees) demonstrates the pattern:

• 7 profiles mapped from サービス開発部 + ML + infra + security + self-improvement
• SOUL templates each reference ミキワメAI products, Rails/Django/React stack, AWS EKS, 5000社データ, NTT東3万名導入
• Kanban board lm-ai-company with swimlanes per profile
• Crons: daily ArXiv scan, hourly health, weekly retro, daily standup digest
• Integration: openmanus_wide_research, company_deepresearch, delegate_task, hyperframes, irodori TTS