Skip to content

docs(feature2): replace CEO mock with real dogfood visual proof of work - #440

Open
LiuShiyuMath wants to merge 1 commit into
mainfrom
worktree-f2-real-visual-proof
Open

docs(feature2): replace CEO mock with real dogfood visual proof of work#440
LiuShiyuMath wants to merge 1 commit into
mainfrom
worktree-f2-real-visual-proof

Conversation

@LiuShiyuMath

Copy link
Copy Markdown
Owner

Summary

Per Stop-hook directive: "fix codes and docs until this html is replaced with real not mocking, tmux + interactive claude x3 + 1 real dashboard, record from real work as demo, implement in a PR and append this html as visual proof of work."

This PR replaces the Feature-2 CEO mock HTML (drafted earlier in /tmp and committed locally as a vision artifact) with a self-contained 626 KB HTML embedding 4 anchors of REAL captured evidence per docs/VISUAL-PROOF-CONTENT.md.

  • docs/mocks/feature2-ceo-live-view.html — visual-proof artifact (replaces 21 KB mock)
  • docs/mocks/_artifacts/demo.cast — asciinema 3.2.0 cast (real)
  • docs/mocks/_artifacts/dashboard-once-stdout.txt — raw pnpm teamagent dashboard --once stdout

Why this PR is unusual

The replaced file was a mock dashboard with fake teammate names and made-up activity. The new file contains only real bytes captured between 15:08-15:22 local on 2026-05-13 from this machine. Nothing on the page is synthesized.

Verification by anyone with this repo checked out

Anchor What was captured How to re-verify
1. tmux + real interactive claude × 3 3 detached tmux sessions running claudefast TUI (Claude Code v2.1.140 + MiniMax-M2.7-highspeed). Sessions cc-alice / cc-bohan / cc-catherine each fed a distinct repo-verifiable prompt. All 3 answered correctly. wc -l docs/POP-OPEN-HTML.md90 (matches alice). ls -d .claude/skills/*/ | wc -l26 (matches bohan). head -12 /Users/m1/projects/AGENTS.md ⇒ matches catherine's reading.
2. real terminal capture asciinema cast committed at docs/mocks/_artifacts/demo.cast. asciinema play docs/mocks/_artifacts/demo.cast replays the full demo.
3. real frontend URL snapshot Headless Chrome render of docs/dashboard.html at 1400×2400 (433 KB PNG, base64-inlined). pnpm teamagent dashboard --once regenerates the live dashboard from real signals.
4. real dashboard raw logs Unfiltered stdout + tail of newest Claude Code project session JSONL (~3 KB). tail -c 4000 ~/.claude/projects/-Users-m1-projects-TeamBrain/*.jsonl | tail -1 shows current real activity.

Honest framing

Feature 2's second-level streaming UX is still vision (per docs/BUSINESS-FEATURES.md 4-layer evidence matrix — Feature #2 turnkey UX is not PRESHIP today). The dashboard rendered in Anchor 3 is the currently-shipping M5 viral-sync surface (hour/day grain). The visual proof demonstrates real activity capture across 3 real Claude instances — it does NOT claim the unfinished streaming layer exists.

Visual proof of work

A reviewer-facing gist + htmlpreview.github.io URL will be appended as a PR comment immediately after creation (per docs/VISUAL-PROOF-FORMAT.md § Hosting recommended-default Gist + htmlpreview path).

Test plan

  • open docs/mocks/feature2-ceo-live-view.html renders without external network (self-contained inline CSS + base64 PNG, no CDN)
  • The 3 <details> tmux blocks contain real Claude Code TUI scrollback (look for MiniMax-M2.7-highspeed banner, real session UUIDs, real wc -l answer = 90, real .claude/skills/ listing)
  • The dashboard screenshot is the real docs/dashboard.html (matches what pnpm teamagent dashboard --once produces locally)
  • htmlpreview.github.io rendering of the gist matches the local render (will be linked in PR comment)
  • /review PASS
  • Human by-hand merge per docs/VISUAL-PROOF-HUMAN-MERGE.md (this PR carries a ## Visual proof of work H2; auto-merge / gh pr merge by agent is blocked by the existing PreToolUse guard)

Per Stop-hook directive ("fix codes and docs until this html is replaced
with real not mocking, tmux + interactive claude x3 + 1 real dashboard,
record from real work as demo, implement in a PR and append this html
as visual proof of work").

Replaces docs/mocks/feature2-ceo-live-view.html (vision mock) with a
self-contained 626 KB HTML embedding 4 anchors of REAL evidence per
docs/VISUAL-PROOF-CONTENT.md:

  1. tmux + real interactive claude × 3
     - 3 detached tmux sessions (cc-alice / cc-bohan / cc-catherine)
     - each running claudefast TUI (Claude Code v2.1.140, MiniMax)
     - each fed a distinct, repo-verifiable prompt
     - all 3 returned correct answers (wc -l, ls .claude/skills/*/,
       read AGENTS.md ASCII art origin)
     - real PIDs + session IDs + tmux list captured at run time

  2. real terminal capture
     - asciinema 3.2.0 cast (docs/mocks/_artifacts/demo.cast, 4.2 KB)
     - demo script captured real ps / tmux list / pnpm teamagent
       dashboard --once / git log / gh pr list output

  3. real frontend URL snapshot
     - headless Chrome render of docs/dashboard.html at 1400x2400
     - 433 KB PNG inlined as base64 data URI
     - dashboard generated from live SQLite + rules + events signals

  4. real dashboard raw logs
     - dashboard-once stdout (incl. SQLite ExperimentalWarning)
     - tail of newest Claude Code session JSONL (~3 KB)

The replaced mock was 21 KB of hand-drawn dashboard with fake teammate
names and made-up activity. The new artifact has only real bytes from
this machine captured between 15:08-15:22 local on 2026-05-13.

Honest framing: Feature 2's second-level streaming UX is still vision.
The dashboard rendered above is the currently-shipping M5 viral-sync
surface (hour/day grain). The visual proof demonstrates real activity
capture across 3 real Claude instances, not the unfinished streaming
layer.

Self-contained per docs/VISUAL-PROOF-FORMAT.md: inline CSS, inline
base64 PNG, no CDN. Renderable via htmlpreview.github.io from the gist
URL appended to PR comments per docs/VISUAL-PROOF-PR.md.
@chatgpt-codex-connector

Copy link
Copy Markdown

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

@LiuShiyuMath

Copy link
Copy Markdown
Owner Author

📎 Visual proof of work · live HTML (hosted on PR proposer's gist per docs/VISUAL-PROOF-FORMAT.md § Hosting)

The 626 KB self-contained HTML committed in this PR (docs/mocks/feature2-ceo-live-view.html) is also published on a public gist owned by the PR proposer (@LiuShiyuMath), reachable from any reviewer's browser:

If htmlpreview shows blank/partial (per the empirical opt-out criteria in VISUAL-PROOF-FORMAT.md § When to prefer self-hosted GH Pages over the Gist + htmlpreview default — heavy inline CSS + 626 KB exceeds 45 KB threshold), use raw.githack as the immediate fallback. The render is byte-identical to opening the file from the repo checkout.

What you should see when the page renders

  • Page title: "TeamBrain · Feature 2 · Real Dogfood Visual Proof of Work"
  • 4 collapsible sections labeled ANCHOR 1-4
  • 3 <details> boxes containing real claudefast TUI scrollback with the MiniMax-M2.7-highspeed banner, real session UUIDs, and verifiable answers (90 / 26 skill names / first 12 lines of AGENTS.md)
  • One large dashboard screenshot rendered from docs/dashboard.html (real M5 viral-sync surface)
  • A compliance table at the bottom marking all 4 anchors PASS
  • Honest framing footer noting Feature 2 second-level streaming is vision (not PRESHIP)

Merge gate

Per docs/VISUAL-PROOF-HUMAN-MERGE.md this PR carries a ## Visual proof of work H2 in the body, so:

  • gh pr merge by agent is BLOCKED by the existing .claude/hooks/visual-proof-merge-guard.sh PreToolUse hook
  • Only a human reviewer pressing the GitHub UI Squash and merge button can land this PR
  • The 6 "solid" checklist items in VISUAL-PROOF-HUMAN-MERGE.md should be inspected before merging

@LiuShiyuMath

Copy link
Copy Markdown
Owner Author

这个html设计不能体现: 1. Alice, Bob这边的状态,Boss那边能直接看到。 请修改。 对应关系在哪?

@LiuShiyuMath

Copy link
Copy Markdown
Owner Author

开始跑feature2 的重新设计了

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant