Repository navigation
feat(chat): Connect buttons for hermes and pi - #683
Merged
Merged
Conversation
omesser
force-pushed
the
feat/675-hermes-pi-connect-buttons
branch
from
September 14, 2026 16:13
a0725ab to
b2037a6
Compare
`harness::launch` names six Harnesses; the landing surface shipped buttons for four, so `hermes` and `pi` were reachable only through Settings. Both now sit in `#landing-buttons` and ride the existing delegated `.connect-btn[data-harness]` handler — no new JS path, and the click still only names the login command. Pi wears Simple Icons' `pi` mark (source `pi.dev`, the Pi coding agent). Hermes is text-only: neither it nor Nous Research has a Simple Icons entry, and the `hermes` slug the README already uses is the German parcel carrier. `tests/connect-buttons.test.js` compares the launch table's arms with the buttons' `data-harness` values, so a seventh named Harness cannot land without one. Closes #675 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The Claude button carried a blocky twelve-point glyph that is not Anthropic's mark. Swap in the real one from Simple Icons. Drop the myHermes parcel-carrier icon from the README's `hermes` row for the same reason, and record the landing in docs/pr-assets. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The button shipped text-only because Simple Icons has no Nous entry and its `hermes` slug is the German parcel carrier. Nous publishes its own monochrome vector at nousresearch.com/safari-pinned-tab.svg, and the same portrait is the favicon on hermes-agent.nousresearch.com. Take that, drop the tracing frame, bake the transform, and ship one path. The README row gets the same mark. It renders through `<img>`, where currentColor has nothing to inherit, so that copy carries an explicit fill instead. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Four of the six connect tiles took a brand colour and two fell through to the neutral default, which left them reading as the unfinished ones. Hermes takes #8C8CF2, Nous's own #0000F2 lifted to carry on the dark Chat ground. Pi takes pi.dev's --color-sunkissed rather than their --color-accent-blue: that blue lands between OpenCode and Hermes and the three then read as one colour. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
The asset was a headless render of chat.html. Replace it with a capture of the built app: launched against a throwaway HOME so no Harness is attached, Summoned by a real double-click on the sprite, captured by window id at 2x. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
omesser
force-pushed
the
feat/675-hermes-pi-connect-buttons
branch
from
September 14, 2026 22:16
b2037a6 to
eea6eea
Compare
omesser
added a commit
that referenced
this pull request
Sep 26, 2026
Three Harness rows had no brand mark and the table read as an oversight. Simple Icons carries no OpenAI, xAI or Block mark, measured against its full 3461-entry index, so the only route is self-hosting. Only xAI grants one. Its brand guidelines permit using the marks "to accurately refer to us or our services", on the condition that the logo is used "exactly as provided" and unaltered, so both official files ship unchanged and `<picture>` picks by theme rather than recolouring either. OpenAI's guidelines say not to use the logo without permission, and Block ships Goose under Apache 2.0, whose section 6 withholds trademark rights. Neither clears the bar, so the README says so instead of leaving a reader to wonder. Attribution was missing for the one mark already self-hosted. `nous.svg` was added in #683 with no record of where it came from; that gap is now written down rather than quietly inherited. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
What changed
src/chat.htmlgains two.connect-btnelements in#landing-buttons, in launch-table order:hermesaftergrok,piafteropencode.tests/connect-buttons.test.jsis new.No JS change. The handler in
src/chat.jsiteratesdocument.querySelectorAll(".connect-btn")and readsbtn.dataset.harness, so both buttons wire themselves. The click still only callsselect_harnessand prints the returned login command — nothing spawns a sign-in, and ai-buddy holds no credential (ADR-0010, #654).src/chat-ui.cssgains the two missing accent blocks..empty .landing-buttonsis a1fr 1frgrid, so six buttons fill three rows. The per-Harness accent blocks are keyed ondata-harness, andhermesandpihad none — they fell through to the neutral default, which left them reading as the unfinished two. Both now carry a brand colour, so all six match.The landing
This is the built app, not a rendered page. It was launched against a throwaway
HOMEso no Harness is attached and the landing is what Chat draws, Summoned by a real double-click on the sprite (verbs: … [Summon]in the log), and captured by window id at 2x. The user's own settings, Memory and Keychain were never read. The file is committed atdocs/pr-assets/chat-landing-connect-buttons.png.Icons
Every mark on the six buttons was checked against the vendor's own artwork.
claudecdn.simpleicons.org/claude(fill#D97757)codexgrok#mark); identical geometryopencodeopencode.ai/favicon.svg— same square-in-square, lightened inner fill for the dark themepipi.dev/favicon.svg, and Simple Icons'pipath verbatim; pi.dev titles itself "Pi Coding Agent", the productpi-acpadapts. Simple Icons is CC0hermesnousresearch.com/safari-pinned-tab.svg, their own monochrome vector; the same portrait is the favicon on nousresearch.com and on hermes-agent.nousresearch.comcdn.simpleicons.org/openaiand/grokboth return 404, so those two paths did not come from Simple Icons. Both were compared directly against the vendors' own published artwork instead, and both match.Drive-by: the Claude button wore a made-up glyph
The
claudebutton shippedM21 10.5h3v3h-3v3h-1.5v3H18…, a blocky twelve-point pixel shape that is not Anthropic's mark. It predates #675. It is fixed here because this PR completes that six-button set, and the same audit that cleared the others caught it — leaving a known-fake vendor logo in a row this PR is finishing would ship the defect knowingly. The real path comes fromcdn.simpleicons.org/claude, inlined withfill="currentColor"and the source's ownviewBox, like its neighbours.README.mdcarried the matching error on itshermesrow:cdn.simpleicons.org/hermesis myHermes, the German parcel carrier. That row now uses the Nous mark, committed atdocs/readme/nous.svgalongside the other README assets. README icons render through<img>, wherecurrentColorhas nothing to inherit and falls back to black, so that copy carries an explicit fill rather thancurrentColor.The Hermes mark: how option A was chosen
Simple Icons has no Nous Research entry —
nous,nousresearchandhermesagentall 404 — and itshermesslug is the parcel carrier. Nous does publish a mark, but it is an illustration rather than a flat glyph, so it was put to the owner as four options:currentColor, same style as the others.The owner picked A, which is what ships. Nothing here is traced or invented — every candidate was the vendor's own file.
Accent colours
--hermes-accent: #8C8CF2is Nous's own#0000F2, which they use across the Hermes surfaces, lifted to carry on the dark Chat ground the way the other accents are.--pi-accent: #E1B06Eis pi.dev's--color-sunkissed. Their own--color-accent-blue(#6A9FCC) was the obvious pick and was rejected on sight: it lands between OpenCode's#7A9EF0and Hermes's#8C8CF2, and the three tiles then read as one colour. Sunkissed is from the same published palette and stays apart — hue 34° against OpenCode's 222° and Hermes's 240°.Why
harness::launchnames six Harnesses. The landing surface offered four, sohermesandpiwere reachable only by opening Settings — Chat's own connect flow covered part of the launch table.The drift test
tests/connect-buttons.test.jsslicesharness.rsfrompub fn launch(to thecustom =>escape hatch, sologin_hint's identically-shaped arms stay out, then compares that set with thedata-harnessvalues inchat.html. It asserts a plausible count on both sides and thatclaudeparsed, so a pattern that stops matching fails rather than passing vacuously. Missing and stray are reported separately, each naming the Harness.How to verify
# tests 125 / # pass 125 / # fail 0.To see the test earn its place, delete the
hermesbutton fromsrc/chat.htmland rerun:Closes #675
🤖 Generated with Claude Code