/search-world

name: search-world
description: "The human needs something that exists somewhere in the world but they don't know where. A decision, a person, a file, a reference — it's been captured, they just can't find it. Searches decisions, people, files, references, insights, and log history across all walnuts in priority order."
user-invocable: true

Find

Search across the world. One verb for all retrieval.

---

How It Searches

Priority order — fastest and highest signal first:

1. Frontmatter Scan (fast, structured)

Scan `_kernel/key.md` across all walnuts. Matches on: type, goal, people names, tags, links, reference descriptions.

2. Insights Search (standing knowledge)

Scan `_kernel/insights.md` across relevant walnuts. Domain knowledge that persists — "Nova Station test windows are Tue-Thu only."

3. Log Search (decisions, history)

Search `_kernel/log.md` entries. Signed decisions, session summaries, what happened when. Frontmatter first (last-entry, summary), then entry bodies.

4. Task Search (work queue)

Search tasks across all walnuts using `tasks.py list` with the `--world` and `--search` flags:

python3 "$ALIVE_PLUGIN_ROOT/scripts/tasks.py" list --world "$WORLD_ROOT" --search "{query}"

Returns JSON with tasks matching the query across all walnuts. Each result includes a `walnut` field for attribution. Combine with `--status`, `--priority`, `--assignee`, `--tag` for narrower results.

5. Working File Search (drafts)

Scan `*/` across walnuts (bundles are flat in walnut root). Find drafts by name, version, age, squirrel attribution.

6. Bundle Manifest Search (captured content metadata)

Search `*/context.manifest.yaml` files across walnuts (bundles are flat in walnut root). Match on frontmatter: type, date, source, participants, subject.

7. Raw Reference Search (expensive)

Load actual raw files. Only on explicit request — "read me that email from Jax."

8. Context Source Cascade (if nothing found locally)

**If steps 1-7 return nothing and the user believes it exists**, automatically fan out to configured context sources before reporting "not found." Check `.alive/preferences.yaml` `context_sources:` for what's available — any configured MCP server, sync script, or API connection is a valid search target.

One-hop inference applies: if the user says "the setup guide I sent Sarah" and an email integration is configured, that's enough to trigger a search without being asked. The system should resolve across its full context surface — local files are not the only source of truth.

> **🐿️ not found in local files**
> Checking configured context sources...
> → Found in email: sent to Sarah, Mar 15, "Setup Guide v2"
>
> ▸ What now?
> 1. Read it
> 2. Capture it to a walnut
> 3. Skip

**Never say "not found" if context sources haven't been checked.**

---

Cross-Walnut Search

Find searches across ALL walnuts by default. Results show which walnut each match came from.

> **🐿️ found 3 matches for "radiation shielding"**
>  1. nova-station / insights.md
>     "Ceramic composites outperform aluminum at 3x the cost"
>
>  2. nova-station / _kernel/log.md — 2026-02-23
>     Decision: go with hybrid shielding approach
>
>  3. nova-station / research/
>     2026-02-23-radiation-shielding-options.md
>
> number to load, or refine search.

Connections

When a match is found, surface connected walnuts:

> **🐿️ [[ryn-okata]] is mentioned in this entry.**
> She also appears in: nova-station, glass-cathedral
> Load her context?

Temporal Queries

"What happened last week" → filter log entries by date range, show across all active walnuts.

"What changed since Tuesday" → scan `_kernel/now.json` updated timestamps + recent log entries.

"History of nova-station" → show `_kernel/log.md` frontmatter (entry count, summary) + offer to load recent entries.