Skip to content
Back to skills

Remnic Search

ASecurity

Run a deep full-text search across every Remnic memory. Trigger phrases include "search memories for", "find anything about", "deep search".

  • 212 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added September 3, 2026
ai-agentsgo

Security analysis

A100/100

Scanned September 3, 2026

npx -y skills add joshuaswarren/remnic --skill remnic-search --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Remnic Search?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Remnic Search
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/joshuaswarren-remnic-search-remnic/badge)](https://www.skillsdirectory.com/skills/joshuaswarren-remnic-search-remnic)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: remnic-search
description: Run a deep full-text search across every Remnic memory. Trigger phrases include "search memories for", "find anything about", "deep search".
allowed-tools:
  - remnic_lcm_search
---

## When to use

Use when a natural-language `remnic-recall` did not surface something the user insists exists, or when the task genuinely needs exhaustive coverage rather than a ranked summary.

Triggers:

- "Search memories for …"
- "Deep search on …"
- "I know I told you about X, find it."
- Follow-up after `remnic-recall` returned nothing useful.

## Inputs

- `query` (required) — literal phrase or keyword string; this is full-text, not semantic.
- Optional: date range, category filter, or entity constraint if the tool supports it.

## Procedure

1. Ask the user (or the calling skill) for the most literal phrase they expect to match.
2. Call `remnic_lcm_search` with that phrase.
3. Group results by date or category when the tool returns enough metadata.
4. Present the top 5–10 matches with dates and short excerpts.
5. If still nothing, say so plainly and suggest `remnic-remember` if the content should be captured going forward.

## Efficiency plan

- Use the most specific phrase available; literal matches are cheaper than wildcarding.
- Prefer one targeted search over several broad ones.
- When recall already worked, do not re-run a deep search for the same topic in the same turn.

## Pitfalls and fixes

- **Pitfall:** Using `remnic_lcm_search` as the default. **Fix:** Start with `remnic_recall`; escalate to search only when recall fails.
- **Pitfall:** Pasting huge result dumps. **Fix:** Show the top matches with excerpts, not the raw payloads.
- **Pitfall:** Over-broad queries like "notes". **Fix:** Require at least one distinctive keyword.

## Verification checklist

- [ ] `remnic_lcm_search` was called only after recall fell short or the task required exhaustive matching.
- [ ] Results were grouped by relevance/date when possible.
- [ ] User-facing output shows excerpts, not raw blobs.
- [ ] Legacy `engram_lcm_search` alias was not preferred over `remnic_lcm_search`.

> Tool names: canonical name is `remnic_lcm_search`. The legacy `engram_lcm_search` alias remains accepted during v1.x.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.