Skip to content

feat: v4.44.0 — deterministic intent routing without slash commands#30

Open
sadewadee wants to merge 1 commit into
mainfrom
feat/auto-intent-routing
Open

feat: v4.44.0 — deterministic intent routing without slash commands#30
sadewadee wants to merge 1 commit into
mainfrom
feat/auto-intent-routing

Conversation

@sadewadee

Copy link
Copy Markdown
Owner

Summary

  • Adds hooks/route-intent.sh wired to UserPromptSubmit: deterministically keyword-matches every prompt for debugging / refine-existing-code / scope-new-feature shapes and injects a hedged suggestion to dispatch dev-squad:qa-engineer / dev-squad:reviewer / dev-squad:architect — no /dev-squad slash command required to engage the swarm.
  • Adds Use PROACTIVELY when... trigger clauses to qa-engineer, reviewer, backend, frontend, architect descriptions so Claude Code's own auto-delegation has a matching path even without the hook (prose-only fires ~50-80% per prior project learnings — the hook is the deterministic backstop, the description edits are the probabilistic fallback).
  • Documents the new hook in the CLAUDE.md hook table.
  • Bumps 4.43.04.44.0 in plugin.json + marketplace.json.

Kept deliberately advisory, not a hard mandate: keyword matching can't distinguish "my app is crashing, fix it" from "explain what this error message generally means", so the injected note always hedges rather than asserting an intent it can't verify. No-ops on prompts starting with / to avoid double-injection on explicit slash-command usage.

Test plan

  • route-intent.sh manually tested against sample prompts (crash/bug, refactor, new-feature, explanatory question, /dev-squad build, unrelated prompt) — fires only on the intended cases, silent otherwise
  • hooks/hooks.json validated as well-formed JSON after edit
  • Install into a live Claude Code session and confirm the injected context appears on a real debugging/refine/feature prompt

Adds a UserPromptSubmit hook (route-intent.sh) that keyword-matches every
prompt for debugging / refine-existing-code / scope-new-feature shapes and
injects a hedged suggestion to dispatch qa-engineer / reviewer / architect —
so the swarm can engage without the user typing /dev-squad first. Prose-only
dispatch (agent description matching) fires ~50-80% of the time per prior
project learnings; this makes routing deterministic while staying advisory,
since keyword matching can't tell "fix this crash" from "explain this error".

Also adds "Use PROACTIVELY when..." trigger clauses to qa-engineer, reviewer,
backend, frontend, and architect descriptions so Claude Code's own dispatcher
has a fallback path even without the hook.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant