Task Types

Different kinds of work have different lifecycle overhead beyond pure coding. The task type multiplier accounts for planning, environment setup, deployment, coordination, and non-coding effort.

Multiplier Table

Type	Multiplier	What It Includes
Coding	1.0×	Baseline — features, refactors, API work
Bug fix	1.2×	Debugging, reproduction, root cause analysis, regression testing
Investigation	0.5×	Timeboxed research spike — output is a plan or report, not code
Design	1.2×	Mockups, iteration with stakeholders, design system alignment
Testing	1.3×	Test environment setup, fixture creation, flakiness debugging, CI integration
Infrastructure	1.5×	Environment provisioning, CI/CD pipeline, deployment verification, monitoring
Data migration	2.0×	Migration planning, schema translation, data validation, rollback strategy, staged rollout

How It's Applied

The multiplier applies to the total estimate (not just agent rounds), because lifecycle overhead affects all phases:

subtotal = agent_time + integration + human_fix + human_review + human_planning
total = subtotal × task_type_multiplier

Auto-Assignment

In quick mode, the skill infers task type from the description:

• "migrate from MySQL to PostgreSQL" → data-migration (2.0×)
• "set up CI/CD pipeline" → infrastructure (1.5×)
• "write e2e tests for checkout" → testing (1.3×)
• "fix login bug" → bug-fix (1.2×)
• "investigate performance issue" → investigation (0.5×)
• "add dark mode toggle" → coding (1.0×)

If the type isn't obvious, the skill asks.

Investigation Tasks

Investigation is special — the 0.5× multiplier makes it smaller than coding. This is intentional:

• Investigations are timeboxed
• The output is a plan, report, or recommendation — not finished code
• The real work happens in a follow-up task estimated separately

Always timebox investigations. A 2-hour investigation spike that reveals a 2-week project is more useful than a 2-week investigation that reveals the same thing.

Calibrating Task Types

If your team's actual overhead for a task type consistently differs from the default multiplier, adjust it. See Calibration for the feedback loop.

Example: If your infrastructure tasks consistently take 2× the estimate, increase the multiplier from 1.5 to 1.8 in your next calibration cycle.