Skip to content
Agent Orchestration
Skill

/vsearch

You CAN search other sessions' conversations. This is a full-text search across every vlx-term session on this machine or within one specified session, so a past discussion is findable even when nobody remembers which session held it. Use it whenever the user gestures at earlier

BOOST
From plugin
velaterm
31027 skills1 MCP
Install
$ npx -y skills add vlinx-io/VelaTerm --skill vsearch --agent claude-code

How it fires

How this skill gets triggered: by you, by Claude, or both.

  • Fires itselfAuto-invocation. Claude auto-loads it when your prompt matches the work.Auto-invocation is when the right skill fires by itself at the right moment, driven by a FLOW.md router and a hook, instead of you invoking it by name. It is the difference between a skill being installed and a skill actually getting used.Read the full definition →
  • You can call itInvoke it directly when you want it.
  • Slash command/vsearch

Context preview

The summary Claude sees to decide when to auto-load this skill.

You CAN search other sessions' conversations. This is a full-text search across every vlx-term session on this machine or within one specified session, so a past discussion is findable even when nobody remembers which session held it. Use it whenever the user gestures at earlier

SKILL.md

vsearch.SKILL.md
name: vsearch
description: >-
  You CAN search other sessions' conversations. This is a full-text search across every vlx-term session
  on this machine or within one specified session, so a past discussion is findable even when nobody remembers which session held it. Use
  it whenever the user gestures at earlier work — "last time", "we solved this before", "the other
  branch", "didn't we discuss this" — in any language, and use it on your own initiative when an error or
  design question feels like one already settled somewhere. Search first, then read the conversation with
  vrefer. Never claim past sessions are unreachable. Read-only: nothing is disturbed. Available only
  inside vlx-term-hosted local sessions.
allowed-tools: Bash(vsearch:*)

vsearch

`vsearch` searches conversation history across **every vlx-term session on this machine**, or within one specified session. It is a pure read: nothing is written, and no other session is interrupted.

When to use it

Reach for it on your own initiative, without waiting to be asked:

  • The user points at another conversation: "how did we fix this last time", "the approach from the other

session", "that decision we made yesterday". The trigger is the meaning, in whatever language it comes.

  • You hit an error, a config value, or an unfamiliar module that feels like it has come up before.
  • You are about to make a design choice that an earlier session may already have settled.

**The failure to avoid:** telling the user that earlier conversations are out of reach. They are not. Search before you say something cannot be found.

Do not use it for searching code or files — that is what Grep and Glob are for. `vsearch` searches **conversations**.

How to run it

vsearch <words...> [--session <session>] [--all|--archived] [--limit N] [--json]
  • Multiple words are an implicit AND and word order does not matter. Two or three distinctive words work

better than a long sentence.

  • `--session <session>` restricts the search to one session. It accepts the same full id, id prefix,

exact name, or unique name substring as `vrefer`; quote names containing spaces. A specified session is searched whether live or archived unless an explicit scope flag is supplied; `--archived` limits it to archived targets, while `--all` retains the broad target scope.

  • By default only live sessions are searched. Add `--all` to include archived ones, or `--archived` to

search only those.

  • `--limit N` caps how many sessions come back (default 10).
  • `--json` gives the full machine-readable result, including every snippet.

What it searches

Conversations only. Sessions whose agent keeps a readable transcript — claude, codex, opencode, pi, omp, and grok — are the ones it can reach. Other agents store their history in private formats that are not parsed yet, so their sessions are invisible to it, as are plain terminal sessions. When a query matched such a session through its terminal recording, the output says how many were left out; that is expected, not a failure.

Reading the output

Each hit shows the session name, the first eight characters of its id, how many matches it has, and up to three snippets. The `msg N` in front of a snippet is that message's index, which `vrefer --range` takes directly. Text output caps the snippets on purpose; use `--json` when you need them all.

Every hit ends with the command that reads that conversation in full:

    full transcript: vrefer a9af7b1c

That is the intended next step — search for the overview, then `vrefer` the session that looks right.

Notes

  • `no matches` on stdout with exit code 0 is a normal answer, not a failure. Say so plainly and move on.
  • The first search after a lot of new conversation activity refreshes the index and can take a few

seconds. It is working, not stuck — do not retry it.

  • What you find comes from **other sessions**, not this one's history. When you use it, say where it came

from.

  • Must run inside a **vlx-term-hosted local session**: it relies on the injected `VLX_*` environment

variables and `vsearch` on PATH. If it reports "not inside a VelaTerm session", the command is simply unavailable here — say so instead of looking for another way in.

Read more
Ships withvelaterm

The best ADE. Not just a terminal. Not just an IDE. Manage agent sessions like Codex. Split terminals like iTerm2. VelaTerm keeps both in one native app, and brings them to your browser and your phone. Watch the intro video (2:55) · Download · User manual

Get the whole plugin
Stats
311
Stars
37
Forks
Active
Maintenance
TypeScript
Language
MIT
License
4d ago
Last commit
2mo ago
Created
7h ago
Added

Repo: vlinx-io/VelaTerm

Other skills on velaterm.