v2.6.0 — Dashboard Elevated: boot sequence, ⌘K palette, tmux bar, mission track

[DanielPodolsky]

Summary

Promotes the approved v3 design experiment into core/templates/dashboard.html.template. The Terminal-Futurism identity (dark, green accent, Outfit + JetBrains Mono, ghost numerals) is preserved — the terminal voice now drives interaction, not just decoration.

The window.PROJECT contract is unchanged. Existing projects upgrade by replacing dashboard.html only; dashboard-data.js is untouched. (Verified live against a real fully-populated project before promotion.)

Added

• Boot sequence — terminal window types live project status on load; any key skips; prefers-reduced-motion bypasses; #noboot disables
• Command palette — ⌘K/Ctrl+K (platform-aware label) or /; digits 1–9 jump views, [ ] cycle
• tmux status bar — live segments: version, project, phase, tasks, DoD %, ticking clock
• Mission track — roadmap as a timeline; progress line interpolates through the active phase's task completion; nodes navigate
• Task burn — per-phase progress bars (hollow for roadmap-only phases)
• Micro-interactions: cursor-glow tiles, count-up stats, kanban column progress, segmented LED DoD bar, marching-ants diagram connectors

Fixed

• Architecture diagram overflow — content-aware layout: label gutter sized to longest label, node boxes sized to their text (computable via JetBrains Mono's fixed metrics), proportional scale-down → font shrink → ellipsis with full text as hover tooltip

Tests

• dashboard-render.test.js: 35 → 48 assertions (every v2.5 contract check kept; mission track, burn, tmux, palette, LED bar added)
• commands.test.sh: +2 v2.6 template-signature checks
• Full suite: 91/91 across 3 layers (bash tests/run.sh)

Version

2.5.0 → 2.6.0 (minor: non-empty Added section, contract unchanged) across installers, templates, contract, README, CHANGELOG.

🤖 Generated with Claude Code

[claude]

Claude finished @DanielPodolsky's task in 3m 32s —— View job

---

Review — v2.6.0 Dashboard Elevated

• Read the diff (git diff origin/main...HEAD)
• Read full context on key files (template, contract, installers, tests)
• Philosophy integrity (P1–P4)
• Command/agent hygiene (S1–S5)
• Cross-platform & migration (T1–T3)
• Security (X1/X2) and docs (D1–D4)
• Posted inline findings + verdict

---

This is a polished, contract-preserving view-layer release. I read the full 977-line template (both render passes), the test changes, and every support-file edit. Two inline comments posted: praise + a defensive NIT on renderArch, and praise on the test discipline.

What's done right

• window.PROJECT contract preserved (T2/T3). dashboard-data.js.template changes only the version string; the schema is untouched. That's what makes "existing projects upgrade by replacing dashboard.html only" actually true — the migration story is correct and is documented in the CHANGELOG. This is exactly the migration awareness T2 asks for, not a casual edit.
• Script parity holds (T1). Both base-install and project-install got the version bump in .sh and .ps1 — no Windows/Mac asymmetry. Manifest version strings updated consistently across init.md, contract, README, and both installers.
• No philosophy regression (P1/P2/P4). This is presentation only — no command flow, no comprehension gate, no profile-affecting branch is touched. Gate 1 is untouched. The terminal voice driving interaction is a UX win, not a comprehension bypass.
• Security hygiene (X1/X2). No secrets. Every user-derived string in the render path flows through esc() before hitting innerHTML/SVG — boot lines, tmux segments, arch labels/nodes, all of it. Install scripts are version-bump-only and keep their quoting.
• renderArch (inline praise). Genuinely well-engineered overflow fix that computes from JetBrains Mono's fixed metrics and degrades gracefully.
• Tests (inline praise). 35 → 48 assertions, every v2.5 contract check kept, async boot made testable via a reduced-motion stub, exact-count assertions guarded by a single-capture trick. The aria-hidden additions on decorative numerals are a real a11y improvement, and the ghost-numeral assertions were correctly updated to regex to match.

Minor / non-blocking

• NIT (D3) — commit scope. release(v2.6.0): — v2.6.0 isn't one of the canonical scopes (commands, install, core, docs, …). You've established a release(vX.Y.Z) convention for release commits and the summary is descriptive and recruiter-skim-able, so the intent of D3 is met — flagging only so the convention stays deliberate.
• NIT — renderArch empty-layers guard (see inline on line 556).
• Observation — mission-track progress line. pct is applied as a width fraction of the node-1→node-N span, so the fill is an aesthetic interpolation rather than landing exactly on node centers proportionally to completed phases. Reads fine; just noting it's approximate, not precise. No action needed.

Caveat: I could not execute bash tests/run.sh / node tests/dashboard-render.test.js in this review sandbox (command approval is disabled here), so the "91/91" claim is not independently re-run by me. I verified the new assertions statically against the render logic and they're consistent — but if you want the green checkmark on record, the CI run on this PR is the source of truth.

Nothing here touches security, philosophy, or installer integrity in a way that blocks merge. Strong, well-tested work that walks its own talk on ownership.

VERDICT: APPROVE
· branch feat/dashboard-design-experiment

[claude]

Claude Code is working…

I'll analyze this and get back to you.

View job run

[claude]

Claude finished @DanielPodolsky's task in 1m 8s —— View job

---

Re-review — new commits on feat/dashboard-design-experiment

• Read the two commits added since my v2.6.0 APPROVE (b5a6159, 31cf88e)
• Verified the renderArch guard against the NIT I raised
• Verified the /own:theme doc-drift fix + CI guard
• Checked for regressions against the verdict

---

Triggered by synchronize. Two commits landed since my approve, and both close real gaps — this is the right way to respond to review.

31cf88e — renderArch empty-layers guard. Resolves the inline NIT from line 552. You went more defensive than I suggested:

if(!d || !Array.isArray(d.layers) || !d.layers.length) return "";

vs. my if(!d.layers?.length). The Array.isArray check also catches the case where layers is a truthy non-array (a malformed data file with layers: {}), which my version would have let through to .map. A bad dashboard-data.js now degrades to an empty diagram instead of corrupt SVG geometry from a -Infinity gutter. Correct fix, placed before any consumer of d.layers. ✅

b5a6159 — /own:theme doc-drift fix (the standout here). This is the kind of fix I like to see called out by name. /own:theme shipped in v2.5.0 but was never added to the command list in core/CLAUDE.md.template — so an agent reading the merged CLAUDE.md in any installed project wouldn't know the command exists (D1 discoverability gap, propagated as a T2 template issue to every install). You didn't just add the missing line — you added a test (commands.test.sh) that walks .claude/commands/own/*.md and asserts each command is documented in the template, so the class of bug fails CI from now on instead of waiting for a human to spot it. Fixing the instance and automating the invariant is exactly the systematic-debugging discipline OwnYourCode preaches. ✅

Minor / non-blocking

• NIT (D3) — commit scope. fix(dashboard): — dashboard still isn't a canonical scope (core would be the home for core/templates/). Same observation as the release(v2.6.0) note on my first pass; the messages are descriptive and recruiter-skimmable, so intent is met. Flagging only to keep the convention deliberate.
• NIT — no regression test on the empty-layers guard. The fix is trivially correct, but a one-line renderArch({layers:[]}) === "" assertion would lock it so a future refactor can't silently reintroduce the -Infinity path. Optional.

Caveat (unchanged): I still can't execute bash tests/run.sh in this sandbox, so the green-checkmark claim rests on CI, not my re-run. The new commands.test.sh assertions and the guard line are statically consistent with the render/contract logic.

Nothing in these two commits touches security, philosophy, or installer integrity. Both are pure improvements over the state I already approved. Verdict stands.

VERDICT: APPROVE
· branch feat/dashboard-design-experiment

[claude]

Claude finished @DanielPodolsky's task in 1m 24s —— View job

---

Re-review — new commit on feat/dashboard-design-experiment

• Read the commit added since my last APPROVE (09d6693)
• Verified the BOM bytes on all three .ps1 scripts (EF BB BF)
• Verified the .sh scripts were not given a BOM (shebang must stay byte-0)
• Verified the new commands.test.sh regression guard
• Checked for regressions against the standing verdict

---

Triggered by synchronize. One commit landed since my last approve, and it's a real-world bug fix done the right way.

09d6693 — UTF-8 BOM on PowerShell scripts (T1, the standout). This is exactly the cross-platform discipline T1 exists to protect. The diagnosis is sharp and correct: PowerShell 5.1 reads a BOM-less file as ANSI, so the UTF-8 banner art decodes to mojibake, and the resulting curly quotes (" ") get accepted by PS 5.1 as string delimiters — which is why the parse errors cascade hundreds of lines away from the actual corruption. The documented fix is a UTF-8 BOM, and that's what you shipped.

What makes this a clean T1 resolution rather than a hack:

• Applied to the right files only. I confirmed the byte prefix on all three: base-install.ps1, project-install.ps1, project-uninstall.ps1 each now start with EF BB BF. Critically, the .sh scripts were left alone — they still start with #! at byte 0. That's the correct nuance most people get wrong: a BOM on a .sh file would break the shebang. Parity means each platform's script is correct for its platform, not byte-identical — and you got that distinction right.
• The instance and the invariant. Like the /own:theme drift guard in b5a6159, you didn't just fix the three files — you added a CI assertion that walks scripts/*.ps1 and fails the build on any BOM-less script. A future-added .ps1 (or a re-save that strips the BOM) now fails CI instead of failing the first Windows user. Fixing the latent bug and automating the guard is the systematic-debugging discipline OwnYourCode preaches.
• Honest provenance. The commit body names it as latent on main since the .ps1 files were created, surfaced during real pre-merge Windows testing — not hidden, not hand-waved. That's ownership over the regression, not just the fix.

Minor / non-blocking

• NIT (D3) — commit scope. fix(windows): — windows still isn't one of the canonical scopes (install is the natural home, since the change is entirely in scripts/). Same note as release(v2.6.0) and fix(dashboard) on my earlier passes: the messages are descriptive and recruiter-skimmable so intent is met — flagging only to keep the convention deliberate. No action required.

Caveat (unchanged): I still can't execute bash tests/run.sh in this sandbox, so the green-checkmark rests on CI, not my re-run. I did independently verify the byte-level claims this commit rests on (od -An -tx1 -N3 → ef bb bf on each .ps1; #! at byte 0 on each .sh), and the new BOM assertion in commands.test.sh is statically consistent with that.

Nothing here touches security, philosophy, or contract integrity. A latent cross-platform bug fixed with a precise diagnosis and a CI guard against recurrence — a pure improvement over the state I already approved. Verdict stands.

VERDICT: APPROVE
· branch feat/dashboard-design-experiment

[claude]

Claude finished @DanielPodolsky's task in 3m 8s —— View job

---

Re-review — new commit on feat/dashboard-design-experiment

• Read the commit added since my last APPROVE (efbd945)
• Verified all three .ps1 are now pure ASCII — grep -P '[^\x00-\x7F]' matches none; no BOM; .sh still #! at byte 0
• Verified the two CLAUDE.md marker regexes still match U+2550 via ═ escapes (semantics byte-for-byte identical to the literal ═ they replaced)
• Verified the flipped test guard (pure-ASCII, stricter than the BOM check it replaces)
• Checked for regressions against the standing verdict

---

Triggered by synchronize. One commit since my last approve — and it's a better fix replacing a fix I already approved, which is the rarest and most valuable kind of follow-up: you went back and questioned your own merged solution.

efbd945 — pure-ASCII .ps1 instead of BOM-stamped (the standout). Last pass I approved 09d6693, which added a UTF-8 BOM to make PS 5.1 read the Unicode banner art correctly. This commit throws that out for something stronger, and the reasoning holds up under inspection:

• The BOM was a partial fix. A BOM makes PS 5.1 decode correctly when the script is run as a file (-File), but OwnYourCode's documented Windows path is irm | iex — piped, not file-loaded — and BOM behavior on that path is ambiguous on PS 5.1. Pure ASCII removes the variable entirely: identical behavior on PS 5.1/7, -File, and irm|iex. That's why the Chocolatey/Scoop installers are pure ASCII, and citing that convention is exactly the evidence-based discipline (P3) the project preaches — you didn't guess, you checked the ecosystem.
• Semantics-preserving where it counts. The two CLAUDE.md cleanup regexes (project-install.ps1:132, project-uninstall.ps1:115) swapped the literal ═ for ═. In .NET regex those are byte-for-byte equivalent — the file becomes ASCII while the pattern still matches the U+2550 box-drawing row written into a merged CLAUDE.md. I diffed against 09d6693 to confirm the only change is the escape form, not the match behavior. Correct call: the marker still resolves on uninstall.
• T1 parity is real, not byte-identical. The .ps1 banner is now ASCII (+===+ / |); the .sh keeps its Unicode box-drawing (base-install.sh:30). That's correct — parity means each platform's script is right for its platform. Bash on macOS/Linux is UTF-8-universal; PS 5.1 is not. Version strings (2.6.0), logic, and flow stay in lockstep across both; only the platform-sensitive glyphs diverge. Same distinction I flagged as correct on the BOM pass, applied consistently here.
• The instance and the invariant, stronger than before. The test guard flipped from "has a BOM prefix" to "has zero non-ASCII bytes" — a strict superset that would have caught the original mojibake bug, not just a stripped BOM. Praise posted inline. This is the third time in this PR you've fixed a bug and automated the guard against its whole class (/own:theme drift, .ps1 encoding ×2). That pattern is the systematic-debugging protocol working as designed.

Minor / non-blocking (unchanged across passes)

• NIT (D3) — commit scope. fix(windows): — windows still isn't a canonical scope (install is the natural home; the change is entirely in scripts/). Messages remain descriptive and recruiter-skimmable, so intent is met — flagging only to keep the convention deliberate. No action required.

Caveat (unchanged): command execution is disabled in this sandbox, so I could not run bash tests/run.sh — the green-checkmark rests on CI, not my re-run. I did independently verify the byte-level claims this commit stands on: grep -P '[^\x00-\x7F]' finds no non-ASCII byte in any of the three .ps1, none carry a BOM, the .sh files still start with #! at byte 0, and the ═ escapes are present in both cleanup regexes. The new ASCII assertion in commands.test.sh is statically consistent with all of that.

Nothing here touches security, philosophy, or contract integrity. A merged fix re-examined, replaced with a more robust one, and locked behind a stricter CI guard — a pure improvement over the state I already approved. Verdict stands.

VERDICT: APPROVE
· branch feat/dashboard-design-experiment

[claude]

Claude encountered an error after 1m 36s —— View job

---

Re-review — new commit on feat/dashboard-design-experiment

• Read the commit added since my last APPROVE (385e350)
• Verified no release was dropped (every version 1.0.0 → 2.6.0 retained)
• Verified the 2.6.0 entry reflects the final pure-ASCII installer state, not the superseded BOM step
• Checked D2 (CHANGELOG discipline) and D3 (commit scope) against the restyle
• Checked for regressions against the standing verdict

---

Triggered by synchronize. One commit since my last approve — a docs-only restyle of CHANGELOG.md to match anthropics/claude-code's convention (version-only headings, flat one-line bullets, no dates/release-names/preamble). 386 deletions, 59 insertions, zero code touched.

385e350 — adopt Claude Code's changelog conventions. This is a clean, evidence-grounded doc change and it holds up under inspection:

• D2 satisfied, semver intact. The 2.6.0 entry is present and the full version ladder is preserved — I diffed the headings old-vs-new and every release 1.0.0→2.6.0 survived ([Unreleased] was empty and the trailing ## Philosophy footer are the only non-version blocks dropped). The "411→110 lines, no release left out" claim in the commit body is true, not hand-waved. Condensing v2.5.0's multi-subsection entry to flat bullets is summarization, not information loss.
• The 2.6.0 entry tells the truth about its own history. Nice catch folding the Windows fix forward: the entry now reads "all .ps1 scripts are now pure ASCII" — the net user-visible state after efbd945 superseded the BOM approach (09d6693). A changelog should describe where the release landed, not narrate the intermediate fix-of-a-fix. Correct call. (P3-adjacent: citing the Chocolatey/Scoop convention in the bullet keeps the claim evidence-tethered rather than asserted.)
• Convention adoption is deliberate, not cargo-culted. Citing anthropics/claude-code as the source convention is the right way to make a style change — it's a real, inspectable precedent, not a personal preference dressed up as a standard.

Minor / non-blocking

• Observation — dropped dates. The previous format carried release dates (## [2.6.0] - 2026-06-11); the new one omits them, matching Claude Code's style. Trade-off worth naming: dates are genuinely useful changelog metadata (Keep-a-Changelog keeps them for a reason — "when did this ship?" is a real reader question). Claude Code drops them and the git tag carries the date, so this is defensible — just flagging it's a real loss, not free.
• Observation — the ## Philosophy footer. The restyle removed the closing manifesto ("If you took away the AI tomorrow, could this junior still code?"). That's brand/mission content, not changelog content, so moving it out of CHANGELOG.md is reasonable — but it's the soul of the product, so make sure it still lives somewhere load-bearing (README / CLAUDE_REVIEWER.md), not just deleted. Non-blocking; just don't let the mission statement fall out of the repo entirely.
• NIT (D3) — commit scope. docs(changelog) — this is the most defensible of the recurring scope notes across this PR: docs is a canonical type, and changelog is a widely-understood conventional scope (the canonical alternative docs(docs) would be sillier). Mentioning only to close the loop on the running D3 thread; no action.

Caveat (unchanged): command execution is disabled in this sandbox, so the 91/91 green-checkmark rests on CI. This commit touches no code or test, so there's nothing test-affecting to re-run regardless — the heading-set diff and entry-accuracy checks above I verified directly against the file and git history.

Nothing here touches security, philosophy enforcement, installer integrity, or the window.PROJECT contract. A documentation hygiene pass that's accurate, loses no release, and corrects its own release note to the final shipped state — a pure improvement over the state I already approved. Verdict stands.

VERDICT: APPROVE
· branch feat/dashboard-design-experiment