← Back to Skills Marketplace
super-memori
by
ciklopentan
· GitHub ↗
· v4.0.23
· MIT-0
647
Downloads
0
Stars
0
Active Installs
68
Versions
Install in OpenClaw
/install super-memori
Description
Local-first hybrid memory skill for OpenClaw agents. Use when the agent needs to find, recall, search, or reuse past knowledge across episodic, semantic, pro...
Usage Guidance
Before installing or enabling this skill, consider the following:
- Review startup-self-check.sh and any '--repair' behavior in detail. The session-start hook's default command includes --repair; decide whether you want automatic repair runs at agent bootstrap. If unsure, disable the hook or remove the --repair flag.
- The auto-learner will read ~/.openclaw/logs/commands.log to propose learning candidates. That file may contain sensitive commands. If you don't want command history scanned, do not enable auto-learner or remove PROMPT_COMMAND logging.
- The hook can inject the generated startup report into the model's bootstrap context. If you need strict separation of local files and model input, disable 'injectBootstrapReport' in the hook configuration before enabling.
- The package uses hardcoded paths (e.g., /home/irtual and absolute npm module paths). Confirm and adjust those paths for your host environment to avoid accidental failures or unintended behavior.
- Run ./health-check.sh --json in a safe/test workspace first to see what the skill would report and whether the host is in degraded mode. Back up your memory workspace (memory/, DB files, and any index-state) before running index/repair operations.
- If you plan to enable the session hook, prefer manual activation and review the ledger/report files it writes (memory/index-state/...). Keep the hook disabled in production until you have tested it in an isolated environment.
Given the documented behaviors and some inconsistencies (repair default vs. 'destructive auto-actions disabled'), treat this skill as potentially privacy-sensitive and proceed only after auditing the scripts and adjusting the hook configuration to match your safety requirements.
Capability Tags
Capability Assessment
Purpose & Capability
The code implements a local-first hybrid memory using local embeddings and Qdrant (127.0.0.1), which aligns with the description. No cloud credentials or remote endpoints are requested. However the package contains a session-start hook and scripts that mutate or repair local state and that read host-specific logs (e.g., ~/.openclaw/logs/commands.log), which are not obvious from a minimal 'memory' description and introduce extra host access.
Instruction Scope
Runtime instructions and code will: run health-checks, query/index/repair local memory, and run a session-start hook that executes ./startup-self-check.sh (the hook's default command includes --repair). The hook can inject the generated report into the agent's bootstrap context (context.bootstrapFiles). The auto-learner reads the user's command log to create 'pending' learning candidates. These behaviors go beyond simple read/query memory operations and can surface sensitive local data into model input and perform repair actions on the host.
Install Mechanism
This is an instruction-only skill (no remote install spec). All code is packaged in the skill (many local scripts). No remote downloads or third‑party install URLs are present in the package metadata, which reduces supply-chain risk. However the hook handler imports OpenClaw internals via absolute paths (e.g., /home/irtual/.npm-global/...), which is brittle and host-specific.
Credentials
The skill requests no environment credentials, which is good, but it reads potentially sensitive local files (command logs, workspace memory dirs, queue/index state) and contacts local services (Qdrant at 127.0.0.1). Reading ~/.openclaw/logs/commands.log and injecting summaries into agent bootstrap context are privacy-relevant and should be justified to users. Hardcoded paths referencing a specific home user ('/home/irtual') are suspicious and may indicate implicit assumptions about host setup or a risk of misconfiguration.
Persistence & Privilege
The skill does not set always:true (good) but includes a hook that — if enabled in the host OpenClaw config — will run at agent:bootstrap and can execute repair logic and write a ledger and a report into the workspace. The hook also injects report content into the model's bootstrap input. Those capabilities give the skill a persistent, automatic touchpoint with the agent lifecycle and the agent's input; combined with automatic repair/run behavior this increases blast radius if misconfigured.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install super-memori - After installation, invoke the skill by name or use
/super-memori - Provide required inputs per the skill's parameter spec and get structured output
Version History
v4.0.23
Add enforceable per-session startup hook via agent:bootstrap + sessionId ledger
v4.0.22
Add startup self-check/self-heal coverage for OpenClaw startup use with honest gateway-start scope and session-start rerun guidance.
v4.0.21
Six-round alternating Dual Thinking hardening pass: release-gate completeness, contract sync, version-truth guardrails, clean-failure fixes, and support-surface clarity improvements.
v4.0.14
Publish 4.0.14: release-surface sync gate, publishedAt validation hardening, workspace-coupled validation disclosure, and path portability fix for validate-release.sh
v4.0.13
Dual Thinking hardening: weak-model routing clarity, publish-honesty fixes, and support-surface truth sync
v4.0.10
Fresh rerun after Dual Thinking hardening; clarified degraded-state routing, degraded miss/partial-result handling, and live-inspection precedence before writes.
v4.0.9
Post-publish truth-sync release: after publishing 4.0.8, updated package/support surfaces to reflect the real published state honestly without changing runtime behavior.
v4.0.8
Six-round alternating Dual Thinking hardening: restored .clawhubignore, advanced honest unpublished line to 4.0.8, synchronized release/support surfaces, added canonical action routing, revalidated strict + weak-model gates.
v4.0.7
Dual Thinking 6-round review: unified Degraded-State Response Catalog (single source for D-WARN-*/D-FAIL); pre-action gate as numbered procedure; Rule 4 flattened to precedence chain; Rule 7 positive memorization test; exit code 0 now handles degraded=true via catalog; top-of-doc status cluster and change-memory sections consolidated; Truth Precedence live-inspection decision rule added; quickstart emits catalog IDs.
v4.0.6
6-round alternating Dual Thinking hardening: support-surface truth sync, weak-model contract cleanup, no runtime capability change
v4.0.5
Restore missing .clawhubignore and republish as a packaging-hygiene fix without changing runtime claims.
v4.0.4
Safe semantic latency fix + final hardening validation
v4.0.3
Benchmark staging hardening, query-anchored semantic rerank, expanded benchmark coverage, refreshed release evidence, and clean strict validation.
v4.0.2
Advance Super Memori to 4.0.2 after a post-publish Dual Thinking documentation/release-surface correction; preserve 4.0.1 validation evidence honestly via inherited-evidence notes instead of mutating the already-published 4.0.1 artifact in place.
v4.0.1
Publish the first registry-honest equipped-host-validated stable line as 4.0.1 after final stable-gate hardening and truth-sync.
v4.0.0-candidate.27
Sync publishedAt metadata to real latest publish state; advance candidate.27 truth surfaces.
v4.0.0-candidate.26
Fix routine maintenance self-noise; reconcile agent-change/hot-buffer tails; refresh candidate.26 validation truth.
v4.0.0-candidate.25
Fix false semantic-health WARN; sync current support surfaces to live candidate.25 host truth.
v4.0.0-candidate.24
Fix canonical indexing so daily notes are indexed and operational memory no longer self-stales freshness; add publish hygiene ignores.
v4.0.0-candidate.23
Extract <details> maintenance block to references/maintenance.md (520→262 lines), fix visibility for weak models, add .clawhubignore, consolidate duplicate exit-code tables
Metadata
Frequently Asked Questions
What is super-memori?
Local-first hybrid memory skill for OpenClaw agents. Use when the agent needs to find, recall, search, or reuse past knowledge across episodic, semantic, pro... It is an AI Agent Skill for Claude Code / OpenClaw, with 647 downloads so far.
How do I install super-memori?
Run "/install super-memori" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is super-memori free?
Yes, super-memori is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does super-memori support?
super-memori is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created super-memori?
It is built and maintained by ciklopentan (@ciklopentan); the current version is v4.0.23.
More Skills