name scry
description Puppet-mode research companion (Hugin): direct research for the project currently being puppeted, save topic summaries, clone reference repos, and hold handed-over artifacts under the target's .puppet/<label>/research/ dir. Read-only by default; meant for use only inside puppet mode. Invoke as /scry.
disable-model-invocation true
allowed-tools Read · Glob · Grep

Scry

Hugin — Odin's raven, thought — flies out, gathers what's needed, and brings it home to roost. /scry is puppet mode's research arm: you direct it, and it gathers knowledge and artifacts into the puppeted target's research/ nest.

Puppet-mode only. /scry is meant to run while you're puppeting a target (after /puppet) — it has nowhere to put what it gathers otherwise. If no target has been adopted this session, stop and warn Brad clearly that /scry is a puppet-mode command, and confirm what he wants before doing anything. There's no enforced mode flag — this is a judgment check, not a hard gate.

Gate — read-only by default. Same stance as /puppet: gather and save into the target's .puppet/<label>/research/ workspace freely, but mutate nothing in either project's real content without an explicit ask. Saving summaries and cloning repos into research/ is scry's whole job and needs no special permission; anything else waits for Brad.

Where things land

Everything scry produces lives under the active target's puppet dir:

  • research/<topic>.md — saved topic summaries.
  • research/repos/<name>/ — cloned reference repositories.
  • research/artifacts/ — files Brad hands over.

Steps

  1. Confirm a target is in play. Identify the project currently being puppeted this session (the one /puppet adopted). If none, give the puppet-mode-only warning above and confirm before continuing.

  2. Do what Brad directs. He drives the research. Common modes:

    • Research a topic → search and read (prefer gh / git clone / raw bytes over WebFetch, per the auto-inherited source-fidelity convention), discuss the findings, and — when he asks — save a summary to research/<topic>.md.
    • Clone a reference repo → issue a single, bare git clone <url> <absolute-dest> into research/repos/<name>/ — never chain it with cd / mkdir / &&. Yggdrasil's permission is a prefix rule (Bash(git clone:*)) that only matches a command line starting with git clone; any wrapper defeats the auto-approve and forces a prompt. git clone already creates the destination and any missing parent dirs, so no mkdir is needed. Gitignored with the rest of .puppet/. Finding the source: the target's own saved references (its References/, any re-pull list, cached files) exist for a reason — check those first. If what you need isn't already on hand, search to resolve the URL (WebSearch, or gh if available) before cloning; don't guess a URL (source-fidelity convention).
    • Take a handed-over artifact → when you're expecting a file from Brad, surface the literal destination path so he can just drop it in — e.g. "Ready for the DAT — toss it in C:\Projects\yggdrasil\.puppet\C-Projects-XI-RE-VibeTesting\research\artifacts\." Then pick it up from there.

  3. Keep the onboarding current. When research produces something worth having next session, add a line to the target's PUPPET.md index so a future /puppet load reflects it.

What this command does NOT do

  • No use outside puppet mode without the explicit warning + Brad's confirmation.
  • No mutation of the target or Yggdrasil's real content — only the .puppet/<label>/research/ workspace.
  • No autonomous research. Brad directs every thread; scry gathers and files, it doesn't decide what to chase.