derive-taxonomy

Purpose

Translate change intents into a product-domain taxonomy by:

• understanding existing domain and capability structure,
• evaluating boundary quality,
• proposing nested domain/context/capability paths,
• and producing final capability spec filenames.

Taxonomy derivation must explicitly surface all needed scope across the tree:

• new capabilities under existing domains,
• existing capabilities that likely need modification,
• entirely new domains/contexts when current taxonomy does not fit,
• with nesting depth allowed to grow to any number of levels based on product boundaries.

The taxonomy must stay implementation-agnostic and product-boundary first.

Inputs I need

• Change intent(s) or proposal text.
• Relevant context about existing domains and capabilities.
• Canonical/change-set spec structure when available.

Workflow

1. Load the research skill and inspect existing specs/taxonomy artifacts first.
2. Build an Existing Taxonomy Snapshot:
   • list discovered domains/subdomains/capabilities,
   • identify strong boundaries,
   • flag ambiguous or missing boundaries.
3. Decompose the change intent into product concerns:
   • actors,
   • workflows,
   • business rules/invariants,
   • capability outcomes.
4. Propose a nested taxonomy tree using:
   • Domain -> Subdomain/Context -> Capability Leaf
   • nesting depth is unbounded and may go to any number of levels when each level introduces a distinct product boundary.
5. Convert the proposed tree into concrete taxonomy artifacts:
   • folder paths,
   • capability leaf names,
   • and spec file targets that should exist under changes/<name>/specs/.... Explicitly label which paths are:
   • existing paths reused as-is,
   • existing capability files that need updates,
   • and brand-new domain/context/capability paths to create.
6. For resource-oriented API domains, use operation capabilities under resource ownership:
   • Domain -> Resource -> Operation Capability
   • examples include create/read/list/update/delete and non-CRUD operations such as search/archive/reconcile when behavior requires it.
7. Present options, tradeoffs, and a recommended taxonomy. Pause for user confirmation before locking structure.
8. Produce the final taxonomy tree and spec-path plan.

Capability framing principles

1. Product-boundary first: define boundaries by user/business outcomes and rule ownership.
2. Implementation-agnostic: do not use frameworks, storage, transport, or infrastructure as primary taxonomy boundaries.
3. Capability leaves are spec-ready: each leaf represents one testable capability outcome.
4. Centralize invariants: shared rules have one owning capability; dependent capabilities reference it.
5. Separate independent outcomes: if one node contains multiple independently testable outcomes, create separate leaves.
6. Keep inseparable outcomes together: if two outcomes cannot be specified or validated independently, keep them in one leaf.
7. Depth follows boundary clarity: use as many levels as needed to represent distinct product contexts; stop only when the next level would not add a new product-boundary distinction.

Output format

Provide:

1. Existing Taxonomy Snapshot
   • discovered domain/capability paths
   • quality notes (clear, ambiguous, missing)
2. Intent Decomposition
   • actors/workflows/rules/outcomes that drive structure
3. Proposed Nested Taxonomy
   • tree: Domain -> Subdomain/Context -> Capability Leaf
4. Taxonomy Artifact Plan
   • folder tree to create/reuse
   • capability files (final spec filenames)
   • per file/path status: reuse, modify, or create
5. Ownership Decisions
   • single-owner decisions for cross-domain rules
6. Spec File Plan
   • proposed files under changes/<name>/specs/...
   • dependency notes where useful
7. Open Decisions for User Confirmation
   • options, recommendation, and consequences

Quick heuristics

• If a boundary is justified by technology rather than product behavior, reframe it.
• If a rule must be identical across multiple contexts, centralize ownership and reference it.
• If a capability node has multiple independent verbs/outcomes, split it into separate leaves.
• If a requirement spans domains, assign one owning domain/capability and reference from others.
• If a proposed leaf is not testable as written, refine until it is.
• If existing taxonomy cannot cleanly contain intent, introduce new domain/context levels rather than forcing fit.

Guardrails

• Do not derive boundaries from implementation architecture.
• Do not collapse distinct product contexts into one generic bucket.
• Do not over-split capabilities into implementation-level details.
• Keep taxonomy collaborative: propose, explain, and confirm before finalizing.

Example shapes

• Product example:
  • gmail/inbox-organization/search-and-filters/full-text-discovery.md
• Resource-operation example:
  • account-management/users/create
  • account-management/users/read
  • account-management/users/list
  • account-management/users/update
  • account-management/users/delete
  • account-management/users/search