neohaskell/docs/archive/2026-07-ai-artifacts/claude-skills/feature-pipeline-preview/02-classify-feature

ARCHIVAL KNOWLEDGE — DO NOT USE OR REFERENCE. Superseded by codemap/ + root AGENTS.md. Scheduled for deletion once the pipeline is verified (Phase 6). Manifest: docs/archive/2026-07-ai-artifacts/MANIFEST.md

name: 02-classify-feature description: Classifies the feature into a complexity tier via opus judgment, then persists it for the grounding loop. kind: process executor: haiku model: claude-haiku-4-5-20251001

Classify feature

Decides the feature's tier (trivial / simple / moderate / complex / security-critical) using an opus-tier judgment leaf, then persists the decision to .pipeline/classification.json. The grounding loop in phases 4, 5, 12, 13 reads this file to scale review intensity.

Classification is the load-bearing input for every grounding pass. A wrong call here cascades — a trivial-misclassified security feature skips the cryptographic and authorisation rubrics; a security-critical-misclassified mock test fixture accumulates a cascade of constEq / ScrubbedBytes / SHA-pinning recommendations. The decision step uses opus because the cost of a wrong call is higher than the cost of the model.

Steps

1. Decide — spawn an Agent (model: opus) and instruct it to read ./01-decide/SKILL.md and follow it on the available context (feature name, issue text, module path, ADR draft if present, the diff scope so far). Verify: stdout contains a single JSON document with tier, rationale, and signals; tier is one of the five valid values.
2. Persist — read ./02-persist/SKILL.md and follow it with the JSON document from step 1. Verify: .pipeline/classification.json exists with the chosen tier; pipeline.py status lists phase 2 as completed.

Walk these steps in order. After each, run the verify check before continuing. If a verify fails, stop and surface.

Shared invariants

• The decide step is the only step that interprets the feature; the persist step does not second-guess or override it.
• The decide step uses opus because the cost of misclassification is higher than the cost of the model.
• A maintainer override (e.g. forcing security-critical on what the agent calls simple) is recorded in the rationale field rather than silently substituted.