Session Compact

Security checks across malware telemetry and agentic risk

Overview

The main session-compaction behavior is coherent, but the package includes an unrelated agent settings file that grants write-capable Git commands, plus it runs shell commands for LLM summaries.

Treat this as a code plugin, not just documentation. Before installing, review or remove the unrelated .qwen permission file, ensure you trust the GitHub/package source, and be aware that auto-compaction stores and reuses conversation summaries locally.

VirusTotal

VirusTotal engine telemetry is currently stale for this artifact.

View on VirusTotal

Risk analysis

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.

#
ASI02: Tool Misuse and Exploitation
Medium
What this means

If an agent honors this settings file after the package is cloned or opened, it may allow repository-changing commands with less review than expected.

Why it was flagged

This dot-directory agent settings file grants write-capable Git commands and WebFetch permissions. Those permissions are not needed for a session-compaction plugin and are not described in SKILL.md.

Skill content
"allow": ["Bash(git log *)", "Bash(git remote *)", "Bash(git push *)", "Bash(git pull *)", "Bash(git add *)", "Bash(git commit *)", "WebFetch(clawhub.ai)"]
Recommendation

Remove .qwen/settings.json from the published package or restrict it to read-only development commands; users should delete or review it before working in the repository with Qwen-compatible agents.

#
ASI05: Unexpected Code Execution
Low
What this means

A malformed or unsafe command construction bug could affect the local environment, even though no unrelated command or exfiltration behavior is shown in the provided artifacts.

Why it was flagged

The static scan shows shell command execution in the compaction engine. This appears tied to the documented OpenClaw CLI LLM-summary workflow, but shell execution is higher risk than a direct API call.

Skill content
const output = execSync(command, {
Recommendation

Review the exact command construction and escaping before installation; prefer a direct OpenClaw API call over shell execution where possible.

#
ASI04: Agentic Supply Chain Vulnerabilities
Info
What this means

Installing this way runs and loads third-party plugin code in the user's OpenClaw environment.

Why it was flagged

The documentation includes a user-directed manual install path that clones source, installs npm dependencies, and builds a local OpenClaw extension. This is normal for a code plugin but is broader than an instruction-only skill.

Skill content
git clone https://github.com/SDC-creator/openclaw-session-compact.git ~/.openclaw/extensions/openclaw-session-compact ... npm install --production ... npm run build
Recommendation

Install only from a trusted source/version, review package contents, and avoid following manual build steps unless you intend to install a code plugin.

#
ASI06: Memory and Context Poisoning
Low
What this means

Private conversation content or summaries may remain on disk and later influence the session context.

Why it was flagged

The plugin stores session messages/summaries locally so conversations can survive restarts and be compacted over time. This is purpose-aligned but involves persistent conversation memory.

Skill content
JSON file-based session storage with version tracking ... Default storage at `~/.openclaw/sessions/`
Recommendation

Use it only for sessions you are comfortable storing locally, review retention/cleanup behavior, and disable auto-compaction for highly sensitive work if needed.