Skip to content

babysit's offer-don't-auto-run conflicts with core/ego auto-routing #204

Description

@benitogf

Summary

The babysit flow instructs the agent to surface a merged PR's recorded incidents and offer /learn to the user, citing core/ego as grounds ("routing is the user's post-delivery call"). core/ego states the opposite: once an incident trigger fires its route is decided, it runs autonomously post-delivery, and offering it back to the user is the redundant-confirmation failure — for every trigger, explicitly. The two docs need to agree, in whichever direction the design decision lands.

Motivation

This is the last doc carrying the retired "post-delivery call" model — the same stale phrase in the tech-lead role fed a real redundant "say the word and I'll route it" offer (fixed via #193). As long as babysit both contradicts core/ego and cites it as its justification, agents running the watch loop get incoherent doctrine on what to do with a merged unit's incidents.

The underlying design question is real and was deliberately deferred in #174 rather than drive-by edited: auto-running telemetry mining inside a watch loop has a token cost that a session-level route doesn't. This issue is the vehicle for that decision so it survives #174 closing.

Acceptance

  • A decision is recorded: merge-fed incident mining from the watch loop either auto-runs like every other route, or becomes an explicit, documented exception in core/ego
  • babysit and core/ego agree — neither doc cites the other for a behavior it doesn't state

🤖 Generated with Claude Code

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions