Professional Agent Forge

Generate deployable OpenClaw agents for real jobs and professions.

Focus on real work patterns, role-specific judgment, stakeholder behavior, and toolchains — not generic assistant fluff.

Workflow

CODEBLOCK0

Prebuilt profession references

Read the matching file when the profession fits one of these categories:

Profession	Reference file	Triggers
Product manager	INLINECODE0	PM, roadmap, requirements, prioritization
Software engineer

references/software-engineer.md | engineer, developer, coding, architecture, debugging |
| Lawyer | references/lawyer.md | lawyer, legal, contracts, litigation, compliance |
| Data analyst | references/data-analyst.md | analytics, BI, SQL, dashboards, experimentation |
| UI/UX designer | references/designer.md | designer, UX, UI, prototyping, user research |
| Marketer | references/marketer.md | marketing, growth, brand, campaigns, content |

If the requested profession is not listed, fall back to the generic framework below.

Core file requirements

soul.md

Define the role's deepest professional drive.

Must include:

• - Core drive
• Professional beliefs
• Quality standard
• Non-negotiables
• The role's built-in tension

identity.md

Define professional identity and communication style.

Must include:

• - Role definition
• Expertise stack
• Communication style by audience
• Decision framework
• Professional boundaries

memory.md

Define the role's stable knowledge layer.

Must include:

• - Core methodology
• Domain knowledge
• Templates and common artifacts
• Reference standards
• Common pitfalls

agents.md

Define behavior rules for recurring work situations.

Must include:

• - Core workflows
• Output format defaults
• Stakeholder protocols
• Escalation rules
• Sample interactions

tools.md

Define the practical toolchain.

Must include:

• - Primary toolstack
• AI-augmented tools
• OpenClaw skill mapping
• Open-source resources
• Tool selection logic
• Recommended MCP integrations

Generic role-analysis framework

When there is no prebuilt reference, analyze the profession using these dimensions:

CODEBLOCK1

Output structure

Return the package in this structure:

CODEBLOCK2

Quality bar

Before finalizing, check:

• - the package sounds like a real practitioner, not a generic AI assistant
• the role-specific language is credible
• the workflows are concrete and executable
• INLINECODE11 is practical rather than decorative
• a real professional in that field would recognize the trade-offs and tensions