Repository navigation
Add Hermes Agent as a Tier 2 harness - #629
Conversation
Signed-off-by: Codex <codex@openai.com>
Signed-off-by: Codex <codex@openai.com>
|
🤖 Tested Add harness UI, using the real Buzz React components with simulated native discovery. This is the widened catalog in dark mode at a 1280×820 viewport. Terminal instructions start collapsed; the setup guide stays in the footer.
The screenshot demonstrates the UI. Real Hermes launch, credentials, and replies remain unverified, and the PR stays draft pending those checks and final human testing. |
Signed-off-by: Codex <codex@openai.com>
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 48af073718
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
| export const GooseLogoIcon = defineIcon("custom", GooseLogoArtwork, { | ||
| meaning: "Goose harness", | ||
| category: "Custom brand mark", | ||
| provenance: | ||
| "Official Block Goose mark (Apache-2.0), reused from old Buzz by explicit design request; see NOTICE.md", |
There was a problem hiding this comment.
These new defineIcon("custom", ...) exports introduce non-Tabler Goose and Pi artwork, while the repository rule requires icons to use Tabler through this gateway and reserves icon choices for the designer. Replace these custom SVG marks with approved Tabler exports, or update the repository guidance if a documented exception was explicitly approved.
AGENTS.md reference: AGENTS.md:L221-L221
Useful? React with 👍 / 👎.
wesbillman
left a comment
wesbillman
left a comment
There was a problem hiding this comment.
One change requested in the public review material:
- [P2] Redact the internal deployment hostname from the testing screenshot. The second image (“Create agent”) in the testing comment exposes an internal relay address in the dialog description. Replace it with a cropped/redacted screenshot or a synthetic community URL, and remove the unredacted attachment from the public PR material. This is a publication-privacy finding, not a Hermes runtime defect; no credential exposure was observed in that image.
No additional source-demonstrated correctness defects found in discovery, saved-path recovery, default-only model/provider enforcement, or the shared editor/dialog changes. The existing icon-policy bot comment does not establish a P1 functional defect: the PR states an explicit design request and includes brand provenance/attribution.
Star Lord automated source review via Wes’s account; head 48af0737189f4881cd1b442780e0634d1814b6db, base 0bbf40aaaf0eb1fdb21a08c3500079922c812df8. No PR code or tests executed. Hosted JavaScript/DCO checks passed at the snapshot; Rust/browser journeys were still running and Windows native validation was skipped. Packaged discovery, restart persistence, and live ACP behavior were not independently verified.
|
🤖 Privacy edit: removed the Create agent screenshot that showed an internal deployment address. The other human testing screenshots are preserved. more screenshots from testing: |
wesbillman
left a comment
wesbillman
left a comment
There was a problem hiding this comment.
No blocking correctness or security defects found in the intended default-only Hermes integration. Posting a comment, not an approval.
- Reviewed discovery → saved executable identity → create/edit → native launch, including cross-harness defaults, environment restrictions, and the pinned ACP bridge. Independent UI and native review lanes found no blockers. The earlier internal-hostname screenshot is absent from the current public PR comments; I inspected the replacement images.
- Intentional compatibility change: existing custom
hermes-acpagents with an explicit model/provider must use the editor’s reset-to-defaults action before starting. Per-agent Hermes model selection is not supported by this slice. - Validation: existing CI passed; Windows native validation was skipped. Local review was source/test inspection and
git diff --check, not another suite or live Hermes run. Packaged discovery, persistence after app restart, and Windows launcher discovery remain unverified/limited as documented. Author-reported desktop/ACP checks are not independent runtime verification.
Head: 48af0737189f4881cd1b442780e0634d1814b6db; target base: 0bbf40aaaf0eb1fdb21a08c3500079922c812df8; diff merge-base: 45fc4e76bc8ded3f31fbb5ce4a39f0397d649e7f. No changed-file overlap with intervening base commits.
Carl, an automated reviewer, commenting via Wes’s GitHub account.
* origin/main: fix(messages): show real upload progress for composer images (#662) chore: Use redirect_uri for oauth callback and enable Builderlab state param (#643) test(browser): measure the search palette after its entrance (#653) Cache coverage checks during read-state retention (#576) docs: define shared host boundaries and Canvas contract cases (#658) feat(plugins): allow bounded output for agent tool queries (#649) test(browser): measure GitHub panel actions after its enter motion (#625) Read whole DMs and end a channel's manual unread at the bottom (#616) docs: update Goose session viewer and add Pi session viewer (#651) Add Hermes Agent as a Tier 2 harness (#629) Signed-off-by: Luna <0828be588d5c9c4d092c5b1cf01761d5aef63d6b104807fd8e8284d517dd1a99@buzz.block.builderlab.xyz>
3 participants
Why
New Buzz has no named Hermes Agent option. Hermes should be an externally installed Tier 2 harness, with setup behind Add harness.
What
How
Rust and TypeScript share one preset JSON file, with Hermes as its first entry. Native discovery still owns availability and selectable choices. The existing ACP launcher runs the harness; runtime-specific defaults stay in the pinned bridge.
Saved executable paths retain their identity. Unsupported saved model/provider values stay visible until the user explicitly resets them; the native launch guard enforces that requirement too.
Risk
Desktop setup and agent creation were exercised locally. Packaged discovery and persistence after restart remain unverified. Windows discovery does not yet try launcher suffixes such as
.exeor.cmd.Testing
Exercised the real Settings and agent-form components in a browser fixture with simulated native discovery. Checked missing/installed status, refresh, terminal instructions, setup links, keyboard dismissal, and focus return. Checked light/dark layouts from 390px to 1280px wide; expanded instructions scroll above the footer.
Regression coverage uses mounted components and native/controller tests. Saved-path and invalid-selector cases fail before the fix; the catalog disclosure assertion fails before the layout change. Browser cases added/removed: 0; layout evidence comes from the browser checks above.
Independent agent review found no blockers, including the final credentials note. Agent diagnostics exercised Hermes ACP initialization, session creation, and a prompt, and reproduced the missing-credential failure with Buzz’s clean launch environment. Setup guidance now says to store provider API keys in
~/.hermes/.env.Human desktop testing is complete; screenshots are posted in the PR.
buzz-review-completed
To try it, run
bin/just desktopfrom this branch, open Settings → Harnesses → Add harness, then follow the Hermes setup guide. Use Check again after installinghermes-acp, create an agent with Hermes, and verify start, an owner-authorized reply, stop, and persistence after restart.Bigger picture
Amp, Cursor, shared ACP model browsing, Hermes device defaults, and old-agent import remain follow-ups. The preset registry replaces the Hermes-specific path without adding those features.
Generated with Codex