Bounded Memory

Provides fast full-text search of OpenClaw session histories with optional LLM summarization for recalling past conversations and decisions.

Security Scan

The skill's code, instructions, and resource access are coherent with its stated purpose (offline indexing/search with optional, opt‑in LLM summaries); there are some minor provenance/documentation inconsistencies you should verify before installing.

Purpose & Capability

The skill claims to index and search OpenClaw session histories and the shipped scripts read ~/.openclaw/agents/*/sessions/*.jsonl, build an SQLite FTS5 DB, and search it. Access to agent session files and an on-disk DB is required and present. The optional LLM summary path that reads an API key from ~/.openclaw/openclaw.json or env vars is consistent with the described opt-in summarization feature.

Instruction Scope

SKILL.md instructs running the included index and search scripts. The scripts only read session files, write a local SQLite DB, and (only when --llm is used) send selected snippets to an external LLM endpoint. There is no instruction to read unrelated system files or exfiltrate data by default. Note: indexing can enumerate all agents when run with --all-agents.

Install Mechanism

No install spec is present (instruction-only); the scripts run as-is with Python and SQLite, which is proportional and low-risk. Nothing is downloaded or extracted by the skill installer.

Credentials

No required env vars are declared. The code optionally reads OPENCLAW_AGENTS_DIR and SESSION_SEARCH_DB_DIR to override paths (reasonable). For summarization it looks for an API key in ~/.openclaw/openclaw.json or env vars (MINIMAX_API_KEY, OPENAI_API_KEY). This is proportional to the optional LLM feature, but be aware it will pick the first provider apiKey present in your OpenClaw config — review that file so you know which key would be used when you enable --llm.

Persistence & Privilege

The skill does not request always:true and runs only when invoked. It writes its own DB file under the skill workspace (or overridden path) and does not modify other skills or global agent settings.

Assessment

This skill appears to do what it claims: index your OpenClaw session .jsonl files into a local SQLite FTS5 DB and search them, with optional LLM summaries only when you pass --llm. Before installing/running: (1) verify the skill source (registry metadata shows unknown/origin inconsistencies in README GitHub links); (2) inspect ~/.openclaw/openclaw.json to see which API keys/providers are stored (the script will use the first apiKey it finds when you enable --llm); (3) only use --llm if you consent to sending matched conversation snippets to the external endpoint (the code calls https://api.minimax.chat/v1); (4) consider where the DB will be stored (SESSION_SEARCH_DB_DIR or the skill db folder) and remove it if you want to revoke the indexed data. If provenance is important, confirm the repository and maintainer identity before installing.

Gives your OpenClaw AI agent a perfect memory — it can recall anything you've ever discussed, even from months ago.

What It Does

Without this skill: each OpenClaw session starts fresh. The AI forgets everything.

With this skill: ask things like:

• "Did we discuss X before?"
• "What did we decide about Y?"
• "Find that conversation from last month"

Privacy Design

What runs	How
Search indexing	✅ 100% offline — SQLite only
Search execution	✅ 100% offline — no network calls
LLM summarization	⚠️ Opt-in only — use --llm flag to enable

No external API calls by default. The --llm flag (disabled by default) sends snippets to your configured LLM for summarization — only when you explicitly ask for it.

Quick Start

# Index your conversations (first time)
python3 skills/session-search/scripts/index-sessions.py --agent main

# Search (fully offline)
python3 skills/session-search/scripts/search-sessions.py "what did we decide about the logo design"

# Search with AI summary (opt-in)
python3 skills/session-search/scripts/search-sessions.py "question" --llm

What It Solves

Problem	Without	With Bounded Memory
"I asked this before but can't remember"	AI has no idea	Instant recall
"What did we decide in that meeting?"	Forgot	Searches all sessions
"Did I mention this before?"	No way to know	Searches everything

Example

You: "Search our conversations about the robot project"
→ Found 3 discussions:
  1. [Last week] We discussed the design direction...
  2. [2 weeks ago] You asked about pricing...
  3. [Last month] The AI suggested adding...