← Back to Skills Marketplace
OpenClaw Memory Canonical
by
ciklopentan
· GitHub ↗
· v4.6.14
· MIT-0
346
Downloads
0
Stars
0
Active Installs
24
Versions
Install in OpenClaw
/install openclaw-memory-canonical
Description
Lightweight file-based memory system for single-user AI agents. Uses markdown files only: HOT/WARM/COLD/BUFFER plus SCRATCH learnings, with tiered loading, W...
Usage Guidance
This skill appears to do what it claims: local file-based memory management with included shell scripts. Before installing or re-syncing into memory/scripts/: (1) review the provided scripts (health-check.sh, atomic-write.sh, archive scripts) to confirm you accept their file operations; (2) back up your current memory/scripts/ and workspace memory/ and .learnings directories because the archive scripts move files and the runtime contract deletes lock files unconditionally; (3) run the health-check from a safe/test copy of your workspace first to observe behavior; and (4) note that no network or credential access is required, so risk is limited to on-disk modifications in the workspace.
Capability Analysis
Type: OpenClaw Skill
Name: openclaw-memory-canonical
Version: 4.6.14
The openclaw-memory-canonical bundle is a comprehensive file-based memory system for AI agents, utilizing Markdown files and local shell scripts for state management. It implements robust patterns such as atomic writes (atomic-write.sh), TTL-based archival (archive-old-episodic.sh), and automated health checks (health-check.sh) to ensure data integrity. While the SKILL.md instructions direct the agent to perform shell operations like file copying, permission changes, and text manipulation (sed), these actions are strictly scoped to the workspace's memory and script directories and are consistent with the stated purpose of maintaining a tiered memory system. No indicators of data exfiltration, unauthorized network access, or malicious persistence were found.
Capability Assessment
Purpose & Capability
Name/description (file-based HOT/WARM/COLD/BUFFER memory) matches the artifacts and runtime instructions. All required artifacts are local files and scripts that operate on the workspace; no unrelated credentials, binaries, or network endpoints are requested.
Instruction Scope
SKILL.md instructs the agent to create/read/update/delete files in the workspace (MEMORY.md, memory/working-buffer.md, .learnings, episodic files) and to copy the packaged scripts into memory/scripts/ and run health-checks. This is expected for a file-based memory system, but pay attention: the decision tree and scripts instruct unconditional deletion of lock files and mv/cp operations (archival moves), which can modify or remove workspace files. Review and back up your workspace/scripts before re-syncing in production.
Install Mechanism
No install spec; this is instruction-only with bundled reference scripts. No network downloads, package-manager installs, or extracted archives are present. The risk surface is limited to the included shell scripts being copied to the workspace.
Credentials
The skill requests no environment variables, no credentials, and no config paths. The scripts operate exclusively on workspace files and directories (memory/, .learnings, MEMORY.md). No secret access or unrelated environment permissions are requested.
Persistence & Privilege
always is false and the skill does not request persistent platform privileges or modify other skills' configs. Its expected persistence is limited to copied runtime scripts under memory/scripts/ and archival of workspace files — normal for a local memory management tool.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install openclaw-memory-canonical - After installation, invoke the skill by name or use
/openclaw-memory-canonical - Provide required inputs per the skill's parameter spec and get structured output
Version History
v4.6.14
Fresh full 10-round Dual Thinking rerun from v4.6.13; no new runtime/code seam found, only honest release-surface refresh for the completed 10-round convergence line.
v4.6.13
Second full 8-round Dual Thinking rerun from 4.6.12 converged with no new material seam; this release is an evidence-refresh / revalidation line, not a functional runtime change.
v4.6.12
8-round alternating Dual Thinking rerun: exact working-buffer header validation in health-check, active-task contract alignment, stale-lock wording alignment, session-end wording alignment, release evidence refresh.
v4.6.11
Fresh 6-round rerun under dual-thinking v8.4.0 strict persistent-chat continuity: tightened package-tree chronology, weekly-review stale-lock cleanup, sequential weekly-review numbering, and exact .pending header validation.
v4.6.10
6-round Dual Thinking rerun: health-check resync-first preflight, deterministic closure-block rotation checks, explicit episodic reassembly branch, and weekly-review single-terminal-closure preservation.
v4.6.9
Fresh 6-round Dual Thinking rerun from 4.6.8: hardened oversized-buffer recovery, clarified packaged-vs-deployed runtime authority, scoped mid-session deferral, made active-subtask counting deterministic, and preserved low-complexity runtime behavior.
v4.6.8
OpenClaw Memory v4.6.8 is a maintenance update focused on documentation quality and evidence traceability.
- Improved clarity, structure, and accuracy across all documentation and reference files.
- Enhanced verification and rationale materials in `references/verification-evidence.md` and `references/design-rationale.md`.
- Refreshed change log in `CHANGELOG.md` to reflect recent updates.
- No behavioral or implementation changes to core memory system logic.
- Existing runtime and execution procedures remain unchanged.
v4.6.7
Add packaged-vs-deployed runtime contract, upgrade resync flow, publish evidence, deterministic package validation, and rollback-safe operator guidance after 6 alternating Dual Thinking rounds.
v4.6.6
Publish hygiene + WARM freshness contract sync + backward-compatible health-check validation
v1.1.1
6-round dual-thinking review; clarified runtime decision tree, health-check gating, heartbeat updates, closure-block preservation, and weak-model execution determinism.
v1.1.0
6 alternating dual-thinking rounds with ai-orchestrator and qwen-orchestrator; tightened weak-model execution, pending recovery, weekly review fallback, and buffer heartbeat invariant
v1.0.4
Fresh 6-round rerun with DeepSeek/Qwen found and fixed two real control-flow issues after the previous release: removed a deadlock where the pre-health-check gate blocked buffer rotation even though rotation is the remediation for the buffer-length invariant, and restored the explicit health-check invocation in the pre-commit gate after that fix. Revalidated successfully in fixture runtime layout.
v1.0.3
6-round dual-thinking pass with DeepSeek/Qwen; fixed portable script root discovery, archive YYYY/MM layout, added legacy learnings archival, aligned health-check with documented invariants, clarified weak-model recovery rules, and removed fragile package-size/file-count claims.
v1.0.2
Clarify Step 9 memory fetch boundary: use positive grep-first loading with explicit stop on zero matches for weaker-model reliability.
v1.0.1
Clarify weak-model execution rules: remove fake retry state, make weekly review candidate detection deterministic, define #pinned syntax, tighten structural write wording, and simplify the material-task-change trigger after dual-thinking review.
v4.6.1
Valid one-chat DeepSeek rerun polish: create MEMORY.md and working-buffer.md if missing at session start, and clarify the exact 4-line working-buffer header skipped by tail -n +5 during rotation.
v4.5.9
DeepSeek-reviewed clarity pass: explicit MEMORY.md header validation, deterministic .pending recovery-failure criteria, semi-autonomous weekly review wording, and runtime-tree subordinate oversized-buffer review check.
v4.5.8
Clarify stale working-buffer recovery, strip working-buffer headers during rotation append, and scope MEMORY.md last_verified guidance for deterministic execution.
v4.5.6
v4.5.6 release after multi-round dual-thinking review with Qwen: removed duplicate startup flow, unified script path convention, simplified weak-model execution rules, tightened buffer rotation/recovery logic, consolidated hard constraints, and improved health-check and promotion-candidate handling.
v4.5.5
Weekly review contradiction resolved; startup recovery priority clarified; scan stop condition added
Metadata
Frequently Asked Questions
What is OpenClaw Memory Canonical?
Lightweight file-based memory system for single-user AI agents. Uses markdown files only: HOT/WARM/COLD/BUFFER plus SCRATCH learnings, with tiered loading, W... It is an AI Agent Skill for Claude Code / OpenClaw, with 346 downloads so far.
How do I install OpenClaw Memory Canonical?
Run "/install openclaw-memory-canonical" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is OpenClaw Memory Canonical free?
Yes, OpenClaw Memory Canonical is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does OpenClaw Memory Canonical support?
OpenClaw Memory Canonical is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created OpenClaw Memory Canonical?
It is built and maintained by ciklopentan (@ciklopentan); the current version is v4.6.14.
More Skills