Skip to content

Task Routes And Lenses

Canonical Keel reference: Task Routes And Lenses.

Rendered directly from the canonical source.

This file contains branch-specific concerns only. SKILL.md owns working-state selection, the constructive loop, Closed/Open completion, and sections 1–8.

Task routes

Choose a route after the working state and initial framing are clear. A route organizes deeper evidence; it does not preselect a solution or require a new artifact. If another route contributes a material concern, import that concern without duplicating the full route.

Route Use when Branch-specific concern
greenfield_design Decision: goals and external constraints exist without inherited repository architecture apply SKILL.md’s greenfield evidence rule; state assumptions, decision authority, and the cheapest way to test the result
architecture_review Judgment: a design, RFC, change set, or existing system needs a load-bearing adequacy, conformance, feasibility, or integration verdict map material claims to evidence or explicit unknowns, preserve focused ownership, and transition to Decision before proposing authorized improvements
boundary_change Decision: a contract, schema, export, flag, entry surface, or compatibility promise is added, widened, moved, or removed identify affected consumers, decision authority, compatibility or cutover commitment, and proportionate verification
structural_refactor Decision: a load-bearing module or seam is split, merged, moved, rewritten, or generalized apply SKILL.md’s Target scope rule; preserve contracts and the deletion path

Complete a selected route only when its branch-specific concern appears in the result or is recorded as an Open unknown. Treat SKILL.md sections 1–8 as the sole source of cross-cutting obligations.

If evidence reopens a materially different direction, return to comparison instead of finishing the current route by inertia.

Rewrite branch

For a structural_refactor that proposes a rewrite, apply SKILL.md’s Target scope rule, derive the target terminal invariant from representative evidence, and start with the smallest independently replaceable implementation slice and its preserved contracts. Expand only when a bounded migration cannot close safely or carries greater evidenced migration or recovery risk.

A whole-target replacement is supported only when the target is bounded; its surfaces, dependents, state, and preserved contracts are known; acceptance evidence maps those contracts and the terminal invariant; runnable rollback or retreat evidence exists; and representative, materially distinct smaller slices do not reach the same invariant at equal or lower evidenced risk. Do not require proof against unbounded hypothetical slices.

Optional review prompts

SKILL.md sections 1–8 are the source of these concerns. Ask only questions whose answers can change the recommendation or a material risk:

  • Which concrete surfaces carry compatibility promises?
  • Which authority, ownership, writer, dependency, or projection relations actually exist?
  • Which negative or time-axis outcome could invalidate the direction?
  • Which guard, migration, recovery, or deletion evidence is necessary at this blast radius?
  • What new concept or layer is added, what does it retire, and is a simpler direction still viable?

Group inapplicable concerns without producing an empty matrix.

Navigation

Type to search…

↑↓ navigate↵ selectEsc close