updated the existing pipeline/v1 schema to support agent properties

[patelraj0602]

Pipeline vs Agent Block Comparison

Overview

This document compares the properties of the Pipeline block and the Agent block in the pipeline.json schema.

Property Comparison

Common Properties (Both Pipeline and Agent)

Both blocks share these core pipeline properties:

• id - Pipeline/Agent identifier
• name - Pipeline/Agent name
• clone - Clone configuration
• inputs - Input variables
• timeout - Execution timeout
• delegate - Delegate selectors
• env - Environment variables
• repo - Repository configuration
• barriers - Pipeline barriers
• if - Conditional execution logic
• on - Trigger configuration
• tags - Pipeline tags
• stages - Pipeline stages array

Pipeline-Specific Properties

Properties only in the Pipeline block:

• allow-stage-executions - Allows stage executions
• fixed-inputs-on-rerun - Fixed inputs on rerun
• notifications - Pipeline notifications

Agent-Specific Properties

Properties only in the Agent block (AI-specific capabilities):

tools (lines 1241-1243)

Configures built-in tools available to all agent steps:

• read - File read capability
• write - File write capability
• grep - Search capability
• bash - Shell command execution

mcp_servers (lines 1245-1251)

Defines Model Context Protocol servers available to all agent steps:

• Command-based (stdio)
• Container-based (Docker)
• HTTP-based (remote)
• Registry-based
• Environment variables
• Allowed tools configuration

rules (lines 1252-1258)

Defines behavioral constraints for the agent:

• Can be file paths (e.g., "./agents.md")
• Can be inline rule text
• Array of strings

skills (lines 1259-1265)

Defines skill files (markdown instructions) for the agent:

• Can be file paths
• Can be inline content
• Array of strings

Design Philosophy

We are trying to keep both agents and pipelines as similar as possible so that we can take maximum advantage of existing pipeline constructs. This means:

• Agents inherit all standard pipeline capabilities
• AI-specific features are additive, not replacement
• Users familiar with pipelines can easily work with agents
• Existing pipeline tooling and infrastructure can be reused

Questions

1. For Rohan

I've tried to keep the pipeline and agent block as similar as possible to leverage the pipeline constructs. Does that sound good?

2. For Harshit

Since we don't have an implementation for top level support of mcp, rules, skills - shall I remove it from the schema as of now?

3. For Rohan

Do we need all different types of stages for starters under the top level Agents construct (line 1222)? Currently the stages array supports:

• UnifiedStageNodeV1
• UnifiedPipelineStageNode
• ParallelStages
• GroupStages
• UnifiedTemplate

Should we simplify this for the initial Agent implementation, or keep all stage types for maximum flexibility?

4. For Harshit

Since we are not using Tools, we shall remove the reference of it, right?

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}