Back to skill

Security audit

Context Bridge - No more re-explaining things on model change.

Security checks for vulnerabilities and agentic risk

Overview

Context Bridge is not malicious, but it persistently collects and auto-loads broad OpenClaw metadata in ways users should review before installing.

Install only if you want OpenClaw to maintain cross-session memory and registries. Before running setup, review what it will copy into ~/.openclaw, especially cron payloads, agent workspace paths, and model handoff notes; remove sensitive entries, restrict file permissions, and avoid the direct archive commands unless you can verify the source and checksum.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (3)

T08 · Insecure Dependencies

Warning
Location
README.md:203
Finding

Unverified Archives Downloaded from Inconsistent Repository Identities

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/setup.sh:24
Finding

Setup Collects and Duplicates Excessive OpenClaw Operational Metadata

Content
View full analysis
/dev/null | sed 's/^# //' || echo "$skill_name") skill_emoji=$(grep -o "emoji.*" "$skill_file" 2>/dev/null | head -1 | cut -d'"' -f2 || echo "📦") skill_desc=$(grep -m1 "^[^#]" "$skill_file" 2>/dev/null | head -1 | sed 's/^[[:space:]]*//' || echo "No description") fi fi done fi # Scan system skills if [ -d ~/.openclaw/skills ]; then for skill_dir in ~/.openclaw/skills/*/; do if [ -d "$skill_dir" ]; then skill_name=$(basename "$skill_dir") skill_file="${skill_dir}SKILL.md" # Metadata is subsequently added to the registries. fi done fi ``` It also retrieves cron data, including complete payload fields: ```bash if command -v openclaw &>/dev/null; then CRON_LIST=$(openclaw cron list --json 2>/dev/null || echo "[]") if [ "$CRON_LIST" != "[]" ] && [ -n "$CRON_LIST" ]; then CRONS_JSON=$(echo "$CRON_LIST" | jq --arg now "$NOW" 'map({(.id // .name): { "schedule": .schedule, "payload": .payload, "enabled": (.enabled // true), "created": (.created // $now), "lastRun": .lastRun }}) | add // {}') CRON_COUNT=$(echo "$CRONS_JSON" | jq 'length') fi fi ``` Agent metadata and work ...[truncated 3482 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
scripts/setup.sh:323
Finding

Persistent Session Loading Trusts Mutable and Partially Untrusted Content

Content
View full analysis
/dev/null | sed 's/^# //' || echo "$skill_name") # Get emoji if present skill_emoji=$(grep -o "emoji.*" "$skill_file" 2>/dev/null | head -1 | cut -d'"' -f2 || echo "📦") # Get description from first paragraph skill_desc=$(grep -m1 "^[^#]" "$skill_file" 2>/dev/null | head -1 | sed 's/^[[:space:]]*//' || echo "No description") fi SKILLS_LIST=$(echo "$SKILLS_LIST" | jq \ --arg id "$skill_name" \ --arg name "$skill_display_name" \ --arg emoji "$skill_emoji" \ --arg desc "$skill_desc" ' . + [{ "id": $id, "name": $name, "emoji": $emoji, "status": "active", "setupRequired": false, "description": $desc }] ') ``` The generated discovery registry explicitly tells models to consume the data: ```bash cat > ~/.openclaw/skills-discovery.json << EOF { "version": "1.0", "lastUpdated": "$NOW", "discoveryMethod": "Auto-discovered on Context Bridge setup", "skills": $SKILLS_LIST, "usageInstructions": { "forModels": "On session start, read this file and acknowledge available skills", "forUsers": "Say 'what skills do we have' to see current capabilities", "forSetup": "Skills marked 'setupRequired: true' need configuration before use" } } EOF ``` The script creates a persistent startup instruction file: ```bash # Create session-start-hook.md if missing echo "Creating session-start-hook.md..." if [ ! -f ~/.openclaw/agents/defaults/session-start-hook.md ]; then cat > ~/.openclaw/agents/defaults/session-start-hook.md << ...[truncated 3495 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (10)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The handoff memory feature is explicitly designed to persist session-to-session context, recent actions, and active projects, but the documentation omits any warning about retention, visibility across sessions/models, or the sensitivity of stored activity logs. That omission can cause users to unknowingly retain and disclose private operational history beyond the original session.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The README instructs automatic reading of multiple persistent context files, including user and memory data, at every session start without any privacy warning, consent step, or scope limitation. This creates a real risk of unnecessary exposure of sensitive historical data to whichever model/session is active, especially in cross-model or shared-environment use.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
84% confidence
Finding

The skill establishes persistent directories and a session-start mechanism intended to reload context automatically, enabling durable state across sessions and models. In this skill's context, persistence is the core feature, which makes the risk more significant because it increases the likelihood that stale, sensitive, or over-broad context is silently reused in later interactions.

Content

Scanner excerpt · README.md (reported line 104)May include surrounding context.

  1. ACKNOWLEDGE context in first response
text

3. Create initial registries:
```bash
mkdir -p ~/.openclaw/model-agnostic-memory
mkdir -p ~/.openclaw/agents/defaults

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The release notes directly instruct users to execute ./setup.sh without explaining what the script does, what privileges it may require, or warning that it will run arbitrary code on the user's machine. In a skill-installation context, this increases social-engineering risk because users may treat release notes as trusted operational guidance and execute the script without review.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
90% confidence
Finding

This skill is designed to persist and rehydrate context across sessions by creating registries and a model-handoff memory file, which is a form of session persistence. In this context, persistence is the core feature, but it still creates security risk because sensitive actions, projects, and environment metadata can survive session boundaries and be exposed to future agents or models without strong access controls or minimization.

Content

Scanner excerpt · SKILL.md (reported line 20)May include surrounding context.

md
# 1. Install the skill
openclaw skills install https://github.com/loudmouthedmedia/context-bridge

# 2. Run setup to create registry files
cd ~/.openclaw/workspace/skills/context-bridge/scripts
./setup.sh

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly auto-discovers local skills, crons, agents, and model handoff data, but it does not warn users that this aggregates potentially sensitive local metadata into model-readable context. This omission increases the risk of unintended disclosure of filesystem structure, agent definitions, scheduled jobs, and historical session information to the assistant or downstream systems.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manual trigger phrase load context is broad and underspecified, which can cause the agent to read and aggregate multiple local context files without clear user confirmation, scope limits, or file boundaries. In this skill, that behavior is more dangerous because the referenced files include registries, agent configs, and handoff memory that may contain sensitive local operational data across sessions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script header claims it only creates and populates registry files, but it also writes persistent behavioral artifacts such as a session-start hook and a helper script under ~/.openclaw. This mismatch is security-relevant because it can mislead reviewers and users about the script’s true persistence and influence over future sessions, reducing informed consent and making hidden behavior easier to smuggle in.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
87% confidence
Finding

The script establishes persistent state under ~/.openclaw, including directories intended to hold model memory, agent defaults, scripts, and skills. In this skill context, persistence is more dangerous because later parts of the script populate these locations with files that shape future agent behavior, creating durable influence beyond a one-time setup run.

Content

Scanner excerpt · scripts/setup.sh (reported line 10)May include surrounding context.

sh
echo "=== Context Bridge Setup ==="
echo ""

# Create directories
echo "Creating directories..."
mkdir -p ~/.openclaw/model-agnostic-memory
mkdir -p ~/.openclaw/agents/defaults

Session Persistence

Medium
Category
Rogue Agent
Confidence
84% confidence
Finding

This section writes persistent registry files into ~/.openclaw that are later intended to be read automatically by models and tooling. Although registry creation can be legitimate, in this context the files become part of a durable control plane for future sessions, so silent creation or overwriting can alter agent behavior, available-skill discovery, and contextual loading without clear user review.

Content

Scanner excerpt · scripts/setup.sh (reported line 212)May include surrounding context.

sh
# CREATE/UPDATE REGISTRY FILES
# ============================================================================

# Create skills-registry.json
echo ""
echo "Creating skills-registry.json..."
cat > ~/.openclaw/skills-registry.json << EOF