| name | routing-flows |
| description | Choose the right task-level flow before doing substantive work, then enter either Superpowers or Humanize intentionally. |
Routing Flows
Overview
Pick the task-level flow first. Then execute that flow consistently.
This is the mandatory entry point whenever a task starts without an active flow, or when new information may invalidate the current flow choice.
If the current work does not already have an explicit active flow, you MUST choose one now.
Do not silently default to Superpowers. Recommend a flow, ask the user to confirm, then continue inside that flow.
If the user already explicitly selected superpowers or humanize for this task, respect that choice.
For repo-changing work, include requesting-code-review and verification-before-completion in Finish from the start. Do not mark the task complete until both have been run.
Repo-changing work includes adding, deleting, generating, formatting, or editing files that belong to the change.
When to Use
Use this whenever a flow must be established or revised.
This includes the start of new work and any change that may invalidate the current task-level flow.
Flow Selection
Recommend superpowers when the task is primarily:
- inspect
- review
- debug
- a bounded config change
- a small implementation that does not need formal plan generation or RLCR
- a task that should use
subagent-driven-development as a side path
Recommend humanize when the task is primarily:
- multi-stage implementation
- cross-cutting work that is expensive to redo
- a task where formal planning already exists and execution should go through Humanize's looped review
- a task where the point is RLCR, hooks, and Humanize state tracking
Shared Planning Backend
brainstorming and Humanize plan generation are shared planning stages.
Use them whenever the task needs an approved design or a formal implementation plan, regardless of whether execution will later continue in superpowers or humanize.
- On Codex, use
humanize-gen-plan.
- On Claude, use
/humanize:gen-plan.
Humanize plan refinement is not a default step after plan generation.
Use it only when the generated plan has been annotated with review comments that need to be resolved and removed before execution.
Decision Style
Before substantive work, publish:
Flow Recommendation: <superpowers|humanize>
Why: <1 short sentence>
Need Confirmation: yes
Then get user confirmation before entering the chosen flow.
Entering Superpowers
If the user confirms superpowers, classify the task on two axes:
task_type: inspect | review | chore | implement | debuglane: Light | Design-only | Designed execution
Choose the most appropriate lane:
Light: bounded analysis, mechanical edits, config work, or small changes with clear scopeDesign-only: design or requirements need clarificationDesigned execution: plan-worthy execution that is not going through Humanize
Then publish:
Flow: superpowers
Route: <task_type>:<lane>
Use: <skills selected for this phase>
Finish: <skills that must be run before the task is complete>
Skip: <heavy skills skipped and why>
Then create a Flow Context. Use the canonical template in references/flow-context.md.
Use lists the skills active right nowFinish lists the skills that must actually be run before the task is complete- for repo-changing work,
Finish MUST include requesting-code-review and verification-before-completion
Entering Humanize
If the user confirms humanize, do not create or propagate a context block.
Instead, hand off directly into the Humanize phase that matches the task state:
brainstorming when design clarification is still needed- plan generation when a formal plan still needs to be generated
- plan refinement only when the plan already exists and contains review annotations
- RLCR when planning is complete and execution should enter the loop
Backend-specific handoff:
- Codex:
humanize-gen-plan, humanize-refine-plan, humanize-rlcr
- Claude:
/humanize:gen-plan, /humanize:refine-plan, /humanize:start-rlcr-loop
Once inside Humanize, follow Humanize's own process and hooks.
Common Mistakes
- defaulting to Superpowers without asking
- treating flow selection and lane selection as the same thing
- treating Humanize plan generation as Humanize-exclusive instead of shared planning infrastructure
- treating plan refinement as mandatory after every plan generation
- generating a context block for Humanize
- leaving review and verification out of Superpowers repo-changing work
- continuing on an old task-level flow after the task has materially changed
