Problem
Operator and agents frequently spawn "future ideas" mid-conversation. Right now those ideas evaporate unless someone manually drafts a plan doc or files an issue (this very issue is one of fourteen filed in one session because the operator dumped a batch).
Proposal
Give the agent a first-class "log a future idea" tool that:
- Accepts a title + body
- Files an issue on the local fork repo (
heavygee/hapi -- never upstream tiann/hapi)
- Labels it
future-idea (new label, needs creating)
- Optionally links to the originating session id
Tool surface:
fileFutureIdea({
title: string,
body: string,
links?: { sessionId?: string, refIssue?: number },
labels?: string[],
})
Authority model:
- Default off; operator enables per-session via a settings flag or per-call with explicit consent.
- Agent must surface the proposed title + body in the response and require operator approval before posting, to avoid noise. Or: post directly and immediately link the issue URL in the response so the operator can close it.
Acceptance
Notes
Problem
Operator and agents frequently spawn "future ideas" mid-conversation. Right now those ideas evaporate unless someone manually drafts a plan doc or files an issue (this very issue is one of fourteen filed in one session because the operator dumped a batch).
Proposal
Give the agent a first-class "log a future idea" tool that:
heavygee/hapi-- never upstreamtiann/hapi)future-idea(new label, needs creating)Tool surface:
Authority model:
Acceptance
heavygee/hapionly (hardcoded -- defence against accidental upstream posts)future-idealabel if missingNotes
messageCodingAgent-- same shape, different sink.