etanhey/phoenix-human-view
skills/golem-powers/phoenix-human-view/SKILL.md
The human-eval UX contract for Phoenix views: turn-by-turn scrollable replay (not a scorecard), hide-but-copyable IDs, collapsed thinking, identity chips, tool filters, tiny frozen starter datasets, mark-wrong-in-thread, mobile-first. Use when: building or reviewing ANY Phoenix/eval view, annotation UI, session replay, or human-grading surface. Triggers: phoenix view, eval UI, annotation view, session replay, human eval UX, grading interface. NOT for: Phoenix data pipelines/ingest (capture scripts have their own specs).
$ install --global
skillsauthnpx skillsauth add etanhey/golems phoenix-human-viewInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
Security Scan Results
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
3
Scanners Passed
9
Scanners in report
Clean
TrivyContainer and dependency vulnerability scanner
95%
Clean
SemgrepStatic code analysis for vulnerabilities
95%
Clean
mcp-scan (Snyk)Model Context Protocol security validation
95%
Skipped
Snyk (dep)Open source security scanning
50%
Skipped
Socket.devSupply chain security analysis
50%
Skipped
VirusTotalMulti-engine malware detection
50%
Skipped
CrowdStrikeAdvanced threat intelligence
50%
Skipped
OSV-ScannerOpen Source Vulnerability database check
50%
Skipped
OWASP Dep-Check
50%
Last scanned: Jun 7, 2026, 3:06 AM11.8s3 files scanned
SKILL.md
- name:
- phoenix-human-view
- description:
- The human-eval UX contract for Phoenix views: turn-by-turn scrollable replay (not a scorecard), hide-but-copyable IDs, collapsed thinking, identity chips, tool filters, tiny frozen starter datasets, mark-wrong-in-thread, mobile-first. Use when: building or reviewing ANY Phoenix/eval view, annotation UI, session replay, or human-grading surface. Triggers: phoenix view, eval UI, annotation view, session replay, human eval UX, grading interface. NOT for: Phoenix data pipelines/ingest (capture scripts have their own specs).
phoenix-human-view — the human-eval UX contract
Documentation-of-standard. This contract was extracted from ~25 live Etan corrections during the gen-10 Phoenix sprint (gen-10 weave #15, imp10) and is ALREADY ENCODED in shipped code (status table below). Any new Phoenix/eval view MUST satisfy it; any review of one checks against it.
The contract
A human-eval view is a reading surface for a human, not a database admin panel.
- Turn-by-turn scrollable REPLAY, not a scorecard. The human reads the conversation as it happened — a CLI-transcript feel — and judges in place. Aggregate scores are secondary chrome, never the primary surface.
- Fewer, human-readable columns. Show what a human reads (who said what, when, verdict). Everything else is detail-on-demand.
- IDs are hidden but copyable. Session/trace/chunk IDs never occupy reading space — but one tap copies them (debugging needs the ID, reading never does).
- Thinking and tool churn collapse by default. Consecutive thinking bundles under one collapsible "💭 Thinking" header, default COLLAPSED, global toggle. Auto-critic flags stay visible even when their turn is collapsed.
- Identity chips: repo + agent + model + role. Every session/turn carries its identity as compact chips — who ran, where, on what model, in which role.
- Tool filters. Filter sessions/turns by which tools were used.
- Tiny frozen starter dataset. Human grading starts on a small FROZEN set (suite-versioned), not a firehose. Frozen = re-gradable = comparable.
- Mark-wrong-in-thread. The human flags a wrong turn WHERE THEY READ IT — in the replay — not in a separate form.
- Mobile-first. Etan grades from his phone at work. Every view ships working mobile layout, screenshot-verified on both (pr-loop Visual Self-QA Gate: clicked-into desktop + mobile shots before merge).
- Turn-type honesty. Local-command artifacts (
<local-command-stdout>,<command-name>, harness caveats) are ⚙️ COMMAND turns, never 🧑 USER turns — mislabeling poisons human judgment of "what the user said."
Status — shipped vs pending (verified 2026-06-05)
Paths below are relative to skills/golem-powers/skill-creator/ — the Phoenix
pipeline lives inside the skill-creator skill, not at repo root.
| Contract item | State | Where |
|---|---|---|
| JSONL→trace ingest (source-of-truth transcripts, not screen scrape) | ✅ #462 | scripts/jsonl_to_phoenix_traces.py |
| Mobile-first annotation view | ✅ #457 | scripts/phoenix_mobile.py + static/phoenix-mobile/ |
| Auto-critic judge write-back + badges | ✅ #458/#459 | scripts/phoenix_auto_critic.py |
| Identity chips (repo/agent/model/role) | ✅ #460 | session cards (scripts/phoenix_mobile.py) |
| Tiny frozen starter dataset (suite-versioned) | ✅ #453-#455 | frozen cmux usage starter set |
| Capture v2 agent identity | ✅ #461 | scripts/cmux_capture.py |
| Project switcher (cmux ↔ coach, one server, :6043 retires) | ⏳ PHX-LEAD | phx-lead-gen2-kickoff.md item 1 |
| Turn-type taxonomy (USER/ORCHESTRATOR/COMMAND/ASSISTANT/TOOL) | ⏳ PHX-LEAD | kickoff item 2 (fixes contract item 10) |
| Thinking-collapse (default collapsed, global toggle) | ⏳ PHX-LEAD | kickoff item 3 |
| Tool-usage filters | ⏳ PHX-LEAD | kickoff item 4 |
| Mobile-from-work persistence (:6042/:6043 → launchd service) | ⏳ open loop | gen-10 weave open-loop #3 — the SAFE always-on (orc C12) |
Update this table when PHX items land — a stale ✅/⏳ here misleads every future view PR.
Hard rules for builders
- Screenshot-gate every view PR: clicked-into desktop + mobile shots → orc → Etan's 👍 before merge (HOLD final design sign-off for Etan — never autonomous).
- Never log raw finding/chunk/personal text into datasets (PII-by-log rule,
eval-harness collab :286). Synthetic docs are QUARANTINED by
suite_version. - /yash-upstream to Arize Phoenix is HOLD-for-Etan — outward-facing.
- Apply
/ui-ux-pro-max,/interaction-design,/html-dashboardfor polish.
Integration
| Skill | Relationship |
|---|---|
| /pr-loop | Visual Self-QA Gate enforces the screenshot-gate on every view PR |
| /never-fabricate | R7 receipts for the visual claims; R9 for dataset counts |
| /skill-creator | Phoenix scripts live in its scripts/; eval datasets follow its RED/GREEN discipline |
| /orc | C12: launchd is the allowed always-on for mobile-from-work persistence |
Related Skills
etanhey/mac-systems
tools
VerifiedTrustedCommunity
macOS systems specialist — AppKit NSPanel architecture, launchd services, socket activation, MCP bridge resilience, syspolicyd, and high-frequency SwiftUI dashboards. Use when building menu-bar apps, LaunchAgents, debugging syspolicyd/Gatekeeper/TCC, resilient UDS/MCP bridges, or SwiftUI dashboards at 10Hz+.
3SKILL.mdUpdated Jun 7, 2026
etanhey/judge-fleet
development
VerifiedTrustedCommunity
Bulk LLM-judging protocol for fleet-dispatched verdict runs (KG cluster, eval harness). Use when: dispatching or running judge workers (J1/J2/RT), planning bulk-apply from verdict JSONL, or triaging evidence_degraded outputs. Triggers: judge fleet, bulk judge, R3 verdicts, kg-judge, RT gate, evidence_degraded. NOT for: single-item code review, Phoenix view UX (use phoenix-human-view), or non-judge eval pipelines.
3SKILL.mdUpdated Jun 7, 2026
etanhey/fleet-wrap
development
VerifiedTrustedCommunity
Quiet-down protocol for sprint close: when the fleet wraps, delete ALL polling crons and monitors, send ONE final dashboard + ONE message, then go SILENT. Use when: fleet wraps, all workers done, overnight queue exhausted, sprint close, Etan asleep/away with nothing approved left. Triggers: fleet wrap, wrap the fleet, stand down, going quiet, sprint close. NOT for: mid-sprint monitoring (keep your loops), spawning a successor (use /session-handoff first).
3SKILL.mdUpdated Jun 7, 2026
etanhey/drive-usage
development
VerifiedTrustedCommunity
Brain Drive filing discipline — where every artifact goes + how to name it. Use WHENEVER touching Google Drive / Brain Drive: uploading, creating folders, saving research prompts/results, audits, plans, transcripts, dashboards, or when about to leave a durable artifact in docs.local/. Teaches the numbered folder model (01_STANDARDS / 02_GROUNDING / 03_RESEARCH / 04_INGEST / 06_ARCHIVE), date-prefixed naming, and the rule: FILE durable artifacts in the right Drive folder — docs.local/ is cache-only. NOT for querying Drive via Gemini (use /braindrive) or web research (use /gemini-research); for >100KB heavy archival defer to /google-drive-archive.
3SKILL.mdUpdated Jun 7, 2026