d2r-parallel-agents

name	d2r-parallel-agents
description	Use when working on multiple D2R characters, data lookups, or validations simultaneously — patterns for dispatching parallel agents on a 16GB Steam Deck

D2R Parallel Agent Dispatch

Overview

Claude Code's Agent tool can parallelize D2R work across characters or tasks. On a 16GB Steam Deck, dispatch at most 2 agents simultaneously to avoid OOM.

When to Parallelize

Good candidates (independent, read-only or isolated writes):

• Validating multiple character YAMLs simultaneously
• Looking up game data for several items at once
• Scanning multiple characters after a batch change
• Reviewing multiple char definitions for issues

Do NOT parallelize:

• Building characters that share stash (collision risk)
• Edits to the same .d2s file
• Anything that writes to shared state

Dispatch Patterns

Pattern 1: Parallel Validation

Validate all characters at once. Each agent validates one character.

Agent({
  description: "Validate Tempest YAML",
  prompt: "Run `python3 -m d2r_chargen validate Tempest` and report results. Only report — do not fix anything."
})
Agent({
  description: "Validate Obsidian YAML",
  prompt: "Run `python3 -m d2r_chargen validate Obsidian` and report results. Only report — do not fix anything."
})

Pattern 2: Parallel Data Lookup

Look up multiple items simultaneously for character planning.

Agent({
  description: "Look up Enigma runeword",
  prompt: "Use the d2r_lookup_runeword MCP tool to look up 'Enigma'. Return the full result including runes, valid bases, and stats."
})
Agent({
  description: "Look up Harlequin Crest",
  prompt: "Use the d2r_lookup_unique MCP tool to look up 'Harlequin Crest'. Return UID, base, and all stats."
})

Pattern 3: Parallel Scan

Scan multiple characters after a mod data change.

Agent({
  description: "Scan Tempest",
  prompt: "Run the d2rdoctor scanner on Tempest. Report: clean, warnings, or hard errors. Full output."
})
Agent({
  description: "Scan Malachar",
  prompt: "Run the d2rdoctor scanner on Malachar. Report: clean, warnings, or hard errors. Full output."
})

Pattern 4: Sequential Build with Parallel Validation

Build characters one at a time (they may share stash), then validate in parallel.

# Step 1: Build sequentially (shared state)
python3 -m d2r_chargen build Tempest
python3 -m d2r_chargen build Obsidian

# Step 2: Scan in parallel (read-only)
Agent({ description: "Scan Tempest", ... })
Agent({ description: "Scan Obsidian", ... })

Constraints

• Max 2 concurrent agents on this Steam Deck (16GB RAM)
• Never run D2R + agents — D2R takes 4-6GB under Proton
• Agents can't share context — each starts fresh, brief them fully
• Read-only agents are always safe — lookups, validation, scanning
• Write agents need coordination — never two agents writing the same file

Agent Prompt Template

When dispatching D2R agents, include:

1. The exact command to run
2. What to report (success/fail, specific fields, full output)
3. Whether to fix issues or just report
4. Relevant rules if the agent will be writing files

Example:

"Run `python3 -m d2r_chargen validate Tempest`.
Report: pass/fail and any validation errors.
Do NOT attempt to fix anything — report only.
If validation passes, also run the d2rdoctor scanner and include its output."