OpenClaw Skill Lazy Loader

PassAudited by ClawScan on May 1, 2026.

Overview

This skill is coherent and low-risk, but it intentionally changes how your agent loads skill and memory context, so review the catalog and AGENTS.md changes before adopting it.

This appears safe to install as an instruction/template skill. Before adopting it, review the AGENTS.md and SKILLS.md changes, keep the skill catalog accurate, and avoid listing sensitive or unrelated files as loadable context.

Findings (3)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

NoteHigh Confidence
ASI01: Agent Goal Hijack
What this means

If the SKILLS.md catalog is incomplete or a task is misclassified, the agent may proceed without relevant skill guidance.

Why it was flagged

This intentionally alters the agent's instruction-loading strategy. It is the skill's stated purpose, but it can affect which skill instructions are available during a task.

Skill content
**At session start:** Read `SKILLS.md` only ... **Do NOT read all skill files at startup.** Load individual skills on demand.
Recommendation

Review the AGENTS.md changes before merging them, keep SKILLS.md complete, and manually tell the agent to load a skill when needed.

What this means

Stale or tampered catalog/memory entries could cause the agent to load the wrong context or expose more local memory context than intended.

Why it was flagged

The skill relies on persistent local catalog and memory files to decide what context the agent loads in future sessions.

Skill content
At session start: Read this file only ... When you need a skill: Read the specific SKILL.md listed below ... `MEMORY.md` | Always
Recommendation

Treat SKILLS.md and MEMORY.md as trusted configuration, restrict who can edit them, and keep listed paths within the intended workspace.

What this means

Running the helper executes local code and reads SKILLS.md from the current directory, but the behavior is disclosed and scoped.

Why it was flagged

The skill includes an optional local Python helper command. The included source shows a simple stdlib-based recommendation tool, not network access or system modification.

Skill content
python3 ~/.openclaw/skills/openclaw-skill-lazy-loader/context_optimizer.py recommend "your next task"
Recommendation

Run the helper only from the intended workspace and review SKILLS.md entries if recommendations look unexpected.