Skip to content

Dynamic Problem Solving

DPS controls how strictly Claude follows the workflow methodology.

LevelNameDescription
0OffExecute exactly as planned. Stop and report any deviation needed.
1MinimalCan ask clarifying questions, minor adjustments within spec.
2LightCan make small fixes/improvements within task scope.
3ModerateCan propose refactors if implementation reveals issues.
4SignificantCan redesign components if plan proves unworkable.
5FullMaximum flexibility. Can pivot approach entirely.

Each command has a sensible default:

CommandDefaultRationale
/research4Exploration needs flexibility
/plan1Should follow research findings
/implement1Should follow the plan
/review3Can propose fixes for gaps
/debug4Root cause analysis is exploratory
/refactor3Can recommend approaches
/archive0Mechanical process

You can override the default when starting a session:

mcp__workflow__start_session({
command: "/implement",
topic: "AUTH-SYSTEM",
dps: 3 // More flexible than default
})
  • Mechanical tasks (archive, migrations)
  • When plan is thoroughly vetted
  • Compliance-sensitive work
  • Implementation of well-defined plans
  • Bug fixes with clear scope
  • Documentation updates
  • Review and gap analysis
  • Refactoring existing code
  • When some exploration is needed
  • Initial research
  • Debugging unknown issues
  • Prototyping and experimentation

When Claude encounters a blocker, the DPS level determines options:

DPS = 0: Must stop and report. Suggest research topics.

DPS > 0: Can propose solutions based on level:

  • Level 1-2: Minor workarounds
  • Level 3: Alternative approaches
  • Level 4-5: Redesigns or pivots
mcp__workflow__log_entry({
tag: "blocker",
fields: {
type: "technical",
title: "API rate limit prevents bulk update",
impact: "Cannot complete migration in one run",
dpsOptions: "- Batch into smaller chunks\n- Add exponential backoff\n- Request rate limit increase",
researchTopics: "- Rate limit best practices\n- Bulk API patterns"
}
})