Repository navigation
server(823): detect matching team skill after send, advise in claim payload - #843
Merged
Merged
Conversation
…ayload Route post-send user messages to matching team skills at chief claim time (worker pre-start routing point): a deterministic matcher (time-expression plus reminder-verb, explicit skill mention; first batch of two rules) appends an advisory-only section to chief.systemPrompt. step.prompt (user text) is never rewritten; ordinary chat and wake rounds get no section, so a misfire costs one ignorable paragraph. The skills inventory in the system prompt now carries descriptions so the agent can verify fit itself. No composer changes: no key conflicts with quoted slash commands or mentions. Evidence: docs/verify/823/ (live-stack probe 11/11 plus send-flow GIF). closes #823
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Implements #823 as corrected: detect the matching team skill after the user sends
(chief claim time = worker pre-start routing point), not while typing.
The reverted #834 ghost-hint direction stays reverted; the composer is untouched,
so there are no key conflicts with
/or@.What changes
packages/shared/src/skill-route.ts(first batch of two rules):explicit mention of an installed skill name (highest precision, wins), and
reminder intent (time expression and reminder verb, both required).
Anything else returns null = no hint.
buildChiefClaimappends an advisory-only section tochief.systemPrompton user-triggered rounds only (wake/system rounds skipped, plus a wake-marker prefix guard in the matcher).
step.prompt(user text) is never rewritten,so a misfire costs one ignorable paragraph and ordinary chat is never hijacked.
so the agent can verify a candidate's fit itself via the
skillstool.Evidence
Live isolated-stack probe
docs/verify/823/probe-chief-skill-route.mjs: 11/11 PASS(
result.json). Claim payload carries the routing section namingmorning-reminderwithset_wakefallback whileinstructionstays verbatim;ordinary chat gets no section; SQLite
step.promptmatches verbatim; drawersend lands in stream and claims with the section.
Checks
pnpm lint: 0 errors;pnpm typecheck: all workspaces pass.packages/shared: 262/262 (incl. 15 new matcher tests).chief,m4a-chief-loop,machine-wire,skills-local,skill-write,chief-skill-route): 124/124.e2e:affected:token-gate.spec.ts4/4 (automatic server-surface fallback).Boundary (honest)
The stack has no live model: this PR proves detection fires and the agent context
is in place, not that a model actually invoked a skill. The "verify before use"
agent side is covered by unit/claim tests plus the standing convention line.
Merging is left to the coordinator.