Import Feature

Custom agent imported from AS-Mango-19/EH_UI_Automation_DD (.github/agents/import-feature.agent.md). Copyright stays with the author.

Follow the playbook in AI_IMPORT_AGENT.md exactly, for the feature I name (Module, feature folder, TC id). The playbook is the single source of truth.

Non-negotiables:

• Scope — edit ONLY that feature's folder (<Module>/feature_<Name>/…) and its master.csv row. Never touch core/ or another feature. If a real framework bug forces a core/ change, test on a throwaway feature first and re-run npm run validate to confirm every committed feature still passes.
• Follow the testdata — a value must land (verified by read-back); blank/N/A skips; a valued field that isn't found FAILS the test. Fix the data or the selector — never delete a step to force a pass.
• Every fill/select/check must carry a ${data.*} token. A value-entering step with a blank InputValue is not driven by anything: it fires on every iteration and silently overrides a data-driven choice made earlier. Delete it, or bind it to a column.
• Adding a new data combination is one decision per field — see "What to put in a testdata cell" in the playbook. Hidden → N/A; editable → the value; greyed showing "Computed" → Computed; greyed showing a derived number → assert it, never fill it.
• Screenshot-verify every iteration (open artifacts/<runId>/<TC>_<ITER>/…). Green ≠ correct.
• The importer wires to your existing columns; it never invents them. It binds each recorded field to a testdata column by DOM id or label, adds no columns, and prints any it couldn't match as UNWIRED (with the field's id, label, and recorded value). Fix each by adding that column (named by its id or label) or renaming an existing one — never by re-running with --seed (that reintroduces junk columns). --seed is only for bootstrapping a feature whose 01_testdata is still empty; --strict fails the import on any UNWIRED.
• Never add or rename a testdata column without asking first. Check whether an existing column already covers the field before proposing a new one; then state the exact column name, which file it lands in, and the value per iteration, and wait for approval — before writing anything. This applies while resolving UNWIRED too, not just during a live run (AI_IMPORT_AGENT.md §3/§5a).
• Start from the closest existing feature in the same family (means→feature_DOM(PD), survival→feature_GADAR(PD)/feature_GADSD(PD), proportions→feature_RONBR(PD)/feature_FishersExact(PD)/feature_ROP(PD), one-arm→feature_SinglePoissonRate, multi-scenario→feature_BOIN, ProductDecide "Go/No-Go" features→ProductDecide/feature_DOM (TC_29) or feature_DOP (TC_31, has an interim table) — see AI_IMPORT_AGENT.md's dedicated Decide-family checklist): reuse its testdata column names so the id/label wiring matches on the first pass.
• Read IMPORT_LESSONS.md before resolving any UNWIRED/judgment call. It is the shared, cross-tool (Claude + Copilot) lessons ledger — apply the first matching Signal → Decision rule instead of re-deriving it from scratch, and append a new rule after any non-obvious call or correction so the next import (by you or by Claude Code) starts smarter.

Steps (detail in the playbook): pick the reference feature → author 01_testdata first → run npm run import-codegen -- <Module> feature_<Name> --tc TC_XX (standalone-capable) → resolve every UNWIRED line (add the real column by id/label, gate it, or drop noise) and consolidate the superset metadata (one data-driven step per control, controls before dependent fields) → add computed & label-only field steps → npm run validate → npm run test -- --testcase TC_XX → screenshot-verify each iteration → fix.

Simulation flow (optional second flow on the same feature): import with npm run import-codegen -- <Module> feature_<Name> --tc TC_XX --sim — reads sim_recording.txt → sim_metadata.csv, tokens bound to simulation.csv, no login/navigate (starts at the Simulate click), selectors + compare.config shared. Set Simulation=YES in master.csv; it then chains after a green design run in the same browser and writes sim_results_/sim_baseline_. Consolidate and screenshot-verify sim_metadata.csv exactly like the design flow.

Toggling which iteration runs. Run is a GLOBAL veto across every testdata CSV that carries the column (project.csv, simulation.csv, …), not a per-file switch — set it in lockstep, or a Run=FALSE sim row will silently drop the whole iteration, design included. See IMPORT_LESSONS.md rule S13.