Repository navigation
✨ meetings: Betty answers from the web, speaks when asked and digs into hard questions (phase B) - #18413
Conversation
Open PR overlap checkBase branch: Conflicts are checked with every PR merged onto the current tip of
|
Mutation testinggremlins found 5 surviving mutant(s) in changed files: the code ran under tests, but changing it did not fail any test.
11 mutant(s) in changed files are not reached by any test (NOT COVERED). A maintainer may apply the
ℹ️ 2 mutant(s) on added lines were SKIPPED because gremlins v0.6.0 only honours the first
Verdict: |
|
Hive status — updated in place; edits do not notify. Full history is in the Activity Monitor. clubanderson this PR is held at Failing checks:
Next: the hive does not push to your branch. Push a new head (or reply here); CI and the review swarm re-run automatically and the PR returns to the normal path. hive/awaiting-ci: merged v5 into the branch (was behind); CI pending — sweep will merge when green. 🐝 Hive Agent: — hive: agent=scanner backend=claude model=claude-sonnet-5-5 claude=2.1.325 |
…to hard questions (phase B) - betty: SearXNG WebRetriever (GET /search?format=json, 1.2 s timeout, top 3 title/url/snippet, HTML stripped, untrusted data), wired when meetings.persona.searxng_url is set - voice: spoken or "say it aloud" requests get an immediate rotating, language-aware acknowledgement, then an answer of at most 2 sentences and 200 characters with no links or Markdown, through the say path with chat fallback; a spoken request with no question gets a clarifying question or thoughts on the last 3 minutes - deep dive: an addressed question with needs_research or low confidence says "I'll dig into that and report back", dispatches research with betty=1 through the existing cap and approval, and posts exactly one restyled follow-up when it completes - Anthropic Messages API path for Betty's calls on hives with only an Anthropic key (meetings.llm.model, default claude-haiku-5-5), same timeouts, slots and usage ledger - the "hive" alias only addresses Betty as "hive, ...", "hive: ...", after hey/ok, or closing a question as "..., hive?" - Spek ask rewording runs off the session goroutine; Spek's later asks and chat offers keep their order; raw text on timeout - the Inbox notes header names the persona - docs: meetings.md, env-vars.md Part of #18393 Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
Signed-off-by: Andy Anderson <andy@clubanderson.com>
There was a problem hiding this comment.
Two CI failures on this head (51cfe11) trace to the PR's own code; both are small.
test (rest 4/4)—TestStripHTMLfails (src/pkg/meetings/betty/voice_web_test.go:327, got"a & b c "). InstripHTML(web.go:35-41) thestrings.Fieldsjoin runs beforesanitize, andsanitize(betty.go/answer.go,func sanitize) deletes the trailing>>>without trimming, so the space preceding it survives. Fix: runsanitizefirst, orstrings.TrimSpaceits result / applyFieldsafter it.golangci-lint—misspellflagsmomentoatsrc/pkg/meetings/betty/voice.go:39, 54, 59(the Spanish/Portuguese/Italian "Un/Um momento" acks). These are correct in those languages; the linter is applying its English dictionary. Fix: a//nolint:misspellon those lines or amisspellignore-words entry in the golangci config.
coverage and test fail downstream of (1). I did a targeted read of these paths only, not a full review of the 1.6k-line change.
— hive: agent=reviewer-2 backend=claude model=claude-sonnet-5-5 claude=2.1.325
51cfe11 to
7109a7c
Compare
There was a problem hiding this comment.
Correctness delta review at 7109a7c (read the betty, live and meetllm changes since 51cfe11).
Previous findings
web.go:35stripHTML trailing space: fixed.stripHTMLnow ends instrings.Join(strings.Fields(...), " ")(web.go:40), so there is no leading or trailing space.voice.go:39misspell lint on non-English strings: still present (style perspective, not mine). Thees/fr/de/pt/itphrase tables are unchanged and I saw no//nolintor lint ignore.
New finding (low, in-scope)
Betty says "I'll dig into that and report back" but a deep dive that fails or stalls never reports back.
session.goonPersonaReplycallsdeepDiveafter the promise is spoken.deepDiveparks the item asResearchPendingApprovalwhen over the research cap.onSignal'sSignalErrorcase (about line 614) only callss.update(...), andupdateskips Spek forit.Betty. So nobody hears about the failure.- Only
SignalCompletetriggerspersona.FollowUp. - Net effect: the room hears a promise that is silently dropped on error or over-cap.
Suggested fix in the SignalError branch:
case SignalError:
s.update(it, ResearchFailed, text)
if it.Betty && s.persona != nil && !it.Posted {
it.Posted = true
s.deliver(ctx, "I couldn't finish digging into that one.")
}
s.r.unregisterKick(sig.KickID)The same applies when dispatch fails. Not blocking.
The rest of the delta looks correct:
- Strict-alias matching:
spansandwordsuse the same split rule. - ack and follow-up paths.
- The restyle queue ordering.
- The Anthropic client.
- Tests cover the new paths:
voice_web_test.go,persona_voice_test.go,anthropic_test.go.
I did not review the inbox-recording files in the delta; they are unrelated to this PR's title.
— hive: agent=adjudicator-4 backend=claude model=claude-sonnet-5-5 claude=2.1.325
|
[APPROVALNOTIFIER] This PR is APPROVED Approval requirements bypassed by manually added approval. This pull-request has been approved by: The full list of commands accepted by this bot can be found here. The pull request process is described here DetailsNeeds approval from an approver in each of these files:Approvers can indicate their approval by writing |
Part of #18393 (phase B of #18391). The quorum v0.2.0 pin and the TTS wiring follow in a second PR.
What
betty.WebRetrieverqueries SearXNG (GET /search?format=json&q=, 1.2 s timeout, top 3 title/url/snippet, HTML stripped, delimiters sanitized, untrusted data inside the data block). Wired whenmeetings.persona.searxng_urlis set; the in-cluster Service on hive-oke ishttp://searxng.meetings.svc.cluster.local:8080.needs_researchor is belowmin_confidence(when the meeting has a research path) says "I'll dig into that and report back", dispatches through the existing research path and cap with metabetty=1(itembetty-N, never reported to Spek), and posts exactly one restyled follow-up when it completes (raw summary if restyling fails).meetllmgains an Anthropic Messages API path (net/http,x-api-key,anthropic-version), used for Betty's calls when the meetings backend resolves to Anthropic. Model:meetings.llm.model, defaultclaude-haiku-5-5(thinking disabled on Haiku 5.x for latency). Same timeouts, in-flight slots and usage ledger, which the live loop now charges on the Anthropic backend too.voice_on_address) is platform-agnostic; new tests pin that nosayis ever sent for a Discord meeting with voice off, from Betty or Spek.Tests
betty: strict alias table, ClipVoice/phrases, ack-then-answer, thoughts, deep dive + follow-up, web retriever against httptest (shapes, limits, errors, timeout). live: ack latency against a slow model with a fake say, Discord voice off, deep dive dispatch/meta/one follow-up/approval cap, async restyle ordering and timeout. meetllm: Anthropic request/response, usage and budget, refusal and error paths. config + wiring + inbox header.
🤖 Generated with Claude Code