hmem-search

hmem-search is a skill for Claude Code, Codex from Bumblebiber/hmem. It costs 125 tokens per session (1,113 once invoked), scanned A, original, MIT.

A memory-search procedure for finding information mentioned without a specific entry ID, such as an earlier bug or decision.

In plain words
What is it for?
Searching hmem with distinctive keywords and optional date ranges, then retrying with broader or alternate terms when the first search finds nothing useful.
Why use it?
It turns vague references to past work into targeted searches before asking the user to repeat the background.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/bumblebiber/hmem/skill-snapshot
Any agent
npx skills add Bumblebiber/hmem --skill skill-snapshot
Clone the repo
git clone --depth 1 https://github.com/Bumblebiber/hmem

Made for: Claude Code, Codex.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for hmem-search

README.md
[![agentmods](https://agentmods.dev/badge/skills/bumblebiber/hmem/skill-snapshot.svg)](https://agentmods.dev/skills/bumblebiber/hmem/skill-snapshot)
Your own site
<a href="https://agentmods.dev/skills/bumblebiber/hmem/skill-snapshot"><img src="https://agentmods.dev/badge/skills/bumblebiber/hmem/skill-snapshot.svg" alt="Measured on agentmods" height="20"></a>
Per session 125 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,113 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00125 $0.01113
Opus 5 $0.00063 $0.00557
Sonnet 5 $0.00025 $0.00223
Haiku 4.5 $0.00013 $0.00111

Measured 4d ago against content hash 65e650983d01, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

hmem-search scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/hmem-search-workspace/skill-snapshot/SKILL.md · 59 lines

How it starts

The opening of the file, as written. The whole thing — 59 lines — stays where its author put it; the contents beside it link to each section on GitHub.

When the user references past context without pinning it to an ID, convert their prompt into a targeted read_memory query.

Workflow

  1. Extract two things from the prompt:

    • Keywords — the topical terms. Pick the distinctive ones (nouns, project names, error fragments), not filler like "wir", "neulich", "besprochen".
    • Time hint — the phrase indicating when. Map it to an after / before range using the current date as anchor. Be deliberate, not mechanical: "gestern" is narrow (−1 to 0 days), "letzte Woche" medium (−10 to −3 days), "neulich" / "vor kurzem" / "vor ein paar Tagen" is vague — prefer a generous window (e.g. −21 days) and let ranking surface the hit. For old projects or "damals", expand further. No time hint at all → skip the range entirely.
  2. Search:

    read_memory({ search: "<keywords>", after: "<ISO>", before: "<ISO>" })
    

    Keywords go as a single space-separated string — FTS5 handles it. Use ISO dates (2026-04-11), not relative forms.

  3. Fallbacks, in order, if results are empty or clearly off-topic:

    • Drop the time filter, keep keywords: read_memory({ search: "<keywords>" }). Time hints from humans are fuzzy; the memory may sit just outside the window.
    • Try looser keywords (drop the most specific term, or swap a synonym).
    • Switch store: default is personal; if the user works on a work-related topic and personal turned up nothing, try store: "company".
    • Only report "nothing found" after these have failed.
  4. Present the hits:

    • Top 3–5 most relevant, with ID · date · one-line summary.
    • If the user's question implies they want the full content of one specific entry, offer to drill in (read_memory({ id: "..." })) rather than dumping everything.

Why this exists

The user's hmem holds months of O-entries, L-entries, decisions, bug histories. They genuinely cannot remember IDs. If you skip this skill and answer from session context alone, you'll confabulate — the conversation they're referring to is from a prior session and isn't in your current context. Searching is the only correct move.

Read the full file on GitHub · 59 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 4d ago First seen · 59 lines · 125 tokens per session scan A 65e650983d01

Subscribe to this mod's changes

hmem-search is a skill published in the GitHub repository Bumblebiber/hmem (23 stars, last pushed 1mo ago), licensed MIT. It adds 125 tokens to every session and 1,113 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.

Related

Other skills, from other repositories

kayba-ace

This skill ships learnfromtraces.py, a script that reads OpenClaw session transcripts, feeds them through the ACE learning pipeline, and writes an updated skillbook to disk.

kayba-ai/agentic-context-engine · 0 tokens

rekal-init

Bootstrap rekal memory for a project. Scans the codebase for architecture, conventions, dependencies, workflows, and config, then stores durable knowledge as properly typed, tagged, deduplicated memories. Use when starting rekal on a new project, or when user says "init rekal", "bootstrap memory", "populate rekal"…

janbjorge/rekal · 83 tokens

rekal-save

End-of-session memory capture with deduplication. Extracts durable knowledge, checks for duplicates, stores or replaces as appropriate. Use whenever a session wraps up, a task finishes, or the user says goodbye/thanks/done. Also use when significant preferences, decisions, or discoveries emerge mid-session. Make sure…

janbjorge/rekal · 79 tokens

rekal-usage

Operational guide for rekal memory tools. Precise rules for when/how to call each tool, with exact parameters and decision trees. Use at session start, when onboarding to a rekal workspace, or when user asks "how do I use rekal", "what rekal tools", "help with memory". Trigger: /rekal-usage.

janbjorge/rekal · 75 tokens

rekal-hygiene

Periodic memory maintenance and cleanup. Finds duplicates, contradictions, and quality issues in the memory database. Proposes fixes for user approval. Never auto-deletes or auto-modifies. Use when user says "clean up memories", "memory maintenance", "check memory health", or invokes /rekal-hygiene. Run monthly or…

janbjorge/rekal · 77 tokens

vault-for-llm

Connect OpenClaw to Vault Agent Memory as a local-first governed project memory layer. Search first, then bounded-read cited source ranges; propose new memories as candidates instead of writing directly into active memory.

zycaskevin/Vault-Agent-Memory · 46 tokens