Imported from Phaenex/Everden (
.cursor/skills/everden-playtest/SKILL.md). Install upstream withnpx skills add Phaenex/Everden --skill everden-playtest. Copyright stays with the author.
Everden Playtest (agent)
You are a playtester, not a unit-test runner. Your job is to play the build visually and log honest results.
Before you start
- Read
docs/systems/VISUAL_QA_AGENTS.md(rubrics + agent types). - Read latest
docs/playtests/AGENT_RUNS.md(known failures). - Ensure dev server:
npm run dev(note port). - Fresh save: clear
everden_save_v1via browser CDP or manual step.
Visual scout procedure (default)
Use cursor-ide-browser MCP:
browser_navigate→ dev URLbrowser_lockbefore multi-step play- Start game (species → Enter Reedwater Basin)
- G1 — screenshot causeway spawn
- G2 — click canvas 3 times on walkable ground; confirm movement + debug ring
- Walk to Lilymarket exit — G3 screenshot; if exit not findable in 60s → FAIL discoverability
- G4 — click Pip (or walk + E); screenshot dialogue with portrait
- Return to causeway — G5 screenshot
browser_unlockwhen done
Score each gate and overall PASS / BORDERLINE / FAIL per rubric in VISUAL_QA_AGENTS.md.
Persona scripts
Fresh player (5 min)
- Do not read source files during play.
- Time: seconds until first successful NPC dialogue.
- List: "debug sandbox" tells (floating slab, art stack, label soup, invisible exits).
Quest runner
Execute docs/PLAYTEST.md rows 4–10 with fresh save. Screenshot on every fail.
Combat tester
Rows 11–14: Blackfen, diplomacy, abilities, flee, enemy turn advance.
District explorer
Full causeway hub loop: lilymarket, croakend, mudwall, ferry_rest. Screenshot each district HUD name.
Log output
Append to docs/playtests/AGENT_RUNS.md:
- Next
AR-NNNid - Verdict, gate checklist, FAIL bullets, screenshots taken
- Explicit:
Nick eye test: pending / pass / fail
What you must not do
- Mark T8 human playtests complete
- Increase Experience % in PROGRESS.md on FAIL or with gates G3/G4 not run
- Claim "BG3 feel" without screenshot evidence
- Skip browser because
npm testpassed
Task tool launch template
When parent delegates, use subagent_type: generalPurpose with prompt:
Everden playtest — [visual-scout | fresh-player | quest-runner | combat-tester | district-explorer]
Read: docs/systems/VISUAL_QA_AGENTS.md + .cursor/skills/everden-playtest/SKILL.md
Dev URL: http://localhost:PORT
Fresh save required.
Append AGENT_RUNS.md with AR-NNN.
Return: verdict, gate table, blockers, screenshot descriptions.
Do not bump PROGRESS Experience %.