The Agent Manifest
Every TACT.md blueprint includes an Agent Manifest — a machine-readable specification that tells any AI system exactly how to build, configure, and run the agent. Think of it as the complete employee file for your AI worker.
Hiring an AI agent is like hiring an employee
When you hire someone, you don't just say "go do stuff." You give them a job description, a code of conduct, you assess their competencies, and you hand them SOPs so they know exactly what to do.
The Agent Manifest does the same thing for AI. Without it, you're just prompting and praying. With it, you have a fully specified, repeatable, auditable AI worker that follows your process — not its own.
“The difference between ChatGPT and an Agent Manifest is the difference between asking a stranger for help and onboarding a trained specialist.”
What's inside an Agent Manifest
DESCRIPTION
Who is this agent, and when does it show up for work?
The Description tells the AI system everything it needs to know about the agent: its name, type (Knowledge, Action, or Decision), what triggers it, which platforms it works on, and a plain-English summary of what it does.
- ✓Agent name and version
- ✓Type classification (Knowledge, Action, Decision)
- ✓Trigger configuration with platform-specific examples
- ✓Tags for discovery and categorisation
- ✓Natural-language description of the agent's purpose
Just like a job description tells you the title, department, reporting line, and responsibilities — the Description tells the AI system exactly what role this agent plays and when it should activate.
RULES
What are the do's and don'ts?
Rules are the guardrails that keep the agent on track. They define what the agent must always do, what it must never do, and the quality standards it must meet. Without rules, an agent is just a chatbot with access to your data.
- ✓Critical "IMPORTANT" and "YOU MUST" directives
- ✓Brand compliance and tone requirements
- ✓Data integrity constraints (no fabrication)
- ✓Format and length standards
- ✓Privacy and confidentiality boundaries
Every organisation has a code of conduct — the non-negotiable behaviours expected of every employee. Rules serve the same purpose for your agent. They ensure it operates within boundaries you define, not boundaries it invents.
SKILLS
What can this agent actually do?
Skills are the specific capabilities the agent can execute — connected to real data sources and tools. Each skill maps to a connector (Google Drive, Salesforce, Slack) and defines when and how it should be used.
- ✓Named skills with clear descriptions
- ✓Mapped connectors per platform (M365 + Google Workspace)
- ✓Trigger conditions: when each skill should activate
- ✓Input/output expectations
When you hire someone, you look at their competencies — can they write SQL? Can they use Salesforce? Can they create a report? Skills are the same thing: they tell the AI exactly what capabilities are available and which tools to use for each.
WORKFLOW
What steps does the agent follow, in what order?
The Workflow is the step-by-step procedure the agent follows from trigger to output. It includes checkpoints for quality control, decision points for branching logic, and a 3-Strike error handling protocol.
- ✓Step-by-step execution sequence
- ✓Checkpoints for quality control
- ✓Skill references at each step
- ✓Error handling with escalation protocol
- ✓Clear outcome definition
A Standard Operating Procedure (SOP) ensures that any employee — new or experienced — can deliver a consistent result. The Workflow does the same for your agent. It doesn't improvise; it follows the procedure you defined.
TACT + Manifest = Your complete agent blueprint
The TACT Overview gives you a human-readable summary. The Agent Manifest gives the AI system a machine-readable specification. Together, they form a complete blueprint that any AI platform can read, understand, and execute.
TACT Overview
For humans. Tells you when it works, who does the work, what data it needs, and what actions it takes.
Agent Manifest
For machines. The full specification that AI systems read to build, deploy, and run the agent automatically.
This isn't just another prompt
Here's what separates a TACT.md blueprint from a ChatGPT prompt.
| ChatGPT Prompt | TACT.md Blueprint | |
|---|---|---|
| Instructions | Ad-hoc, varied | Structured Agent Manifest |
| Guardrails | None | RULES with critical directives |
| Data access | Copy-paste only | Connected via SKILLS & connectors |
| Process | Improvises each time | Follows a defined WORKFLOW |
| Repeatability | Different every time | Consistent, auditable |
| Portability | Locked to one chat | Works across 4+ platforms |
See the Agent Manifest in action
Every blueprint in our library includes a full Agent Manifest. Browse, download, and deploy your first agent today.