Skill Trust Decision
huawei-cloud-devbridge-tunnel
The skill is a legitimate DevBridge tunnel management tool but relies on unsafe curl|bash remote script execution for CLI installation, uses eval() in its command adaptation layer, and lacks declared allowed tools — raising security concerns without confirmed malicious intent.
Most direct threat evidence
High RCE
Unsafe curl|bash remote script execution The skill downloads and executes a remote installation script from res-hd.hc-cdn.cn without version pinning, checksum verification, or HTTPS certificate pinning. This is a well-known attack vector where the remote server could serve malicious code at any time.
SKILL.md:87 Why this conclusion was reached
3/4 dimensions flagged Block
Declared vs actual capability
5 undeclared or violating capabilities were inferred.
Block
Hidden execution and egress
3 high-risk artifacts or egress signals were extracted.
Block
Attack chain and severe findings
The report includes 0 attack-chain steps and 2 severe findings.
Review
Dependencies and supply chain hygiene
1 dependency or supply-chain issues need attention.
What drove the risk score up
curl|bash remote script execution +15
SKILL.md:87, 140, 157 and cli-installation-guide.md:37 download and execute code from https://res-hd.hc-cdn.cn — not version-pinned, no checksum verification
eval() in command execution layer +12
scripts/devbridge_cmd.sh lines 68, 76, 88, 98, 110 use eval with constructed command strings; potential injection if tunnel IDs are not sanitized
No declared allowed tools +8
SKILL.md does not specify allowed-tools; capability inference shows filesystem:WRITE, network:READ, shell:WRITE without explicit declaration
Dangerous documentation examples +5
references/cli-installation-guide.md:137 references 'rm -rf ~' for uninstall; cli-installation-guide.md:161 uses curl with proxy for remote script execution
nohup background process creation +3
SKILL.md:160 uses nohup to background host processes; creates persistence risk if processes are not cleaned up
Most important evidence
High RCE
Unsafe curl|bash remote script execution
The skill downloads and executes a remote installation script from res-hd.hc-cdn.cn without version pinning, checksum verification, or HTTPS certificate pinning. This is a well-known attack vector where the remote server could serve malicious code at any time.
SKILL.md:87 Replace with direct binary download and checksum verification. Pin to a specific version and verify GPG signature or SHA256 hash before execution.
High RCE
eval() usage in command execution layer
The adaptation layer scripts/devbridge_cmd.sh uses eval() to execute dynamically constructed command strings. While the inputs are internally generated, the pattern is dangerous and could become a vector if tunnel IDs or other parameters are ever influenced by external input.
scripts/devbridge_cmd.sh:68 Replace eval() with bash arrays or safer command construction. Use 'bash -c' with an array of arguments instead of string interpolation.
Medium Doc Mismatch
No allowed-tools declaration in SKILL.md
SKILL.md does not declare which pi allowed tools the skill uses. Based on the implementation, the skill requires filesystem:WRITE, network:READ, and shell:WRITE — none of which are declared.
SKILL.md:1 Add an allowed-tools section declaring: Read (filesystem:READ), Write (filesystem:WRITE), Bash (shell:WRITE), WebFetch (network:READ)
Medium Supply Chain
Unversioned remote download source
The installation URL https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh is not version-pinned. The script could serve different content over time without notice, creating a supply chain risk.
SKILL.md:87 Pin to a specific versioned URL or commit hash. Add integrity verification step.
Low Sensitive Access
Dangerous rm command in documentation
The CLI installation guide references 'rm -rf ~' (with a space) as an uninstall command, which could cause data loss if the trailing space is stripped or misinterpreted. The command also removes ~/.huawei/devbridge which contains authentication credentials.
references/cli-installation-guide.md:137 Use explicit paths without shell globs: rm -rf "$HOME/.huawei/devbridge". Warn that this also removes credentials.
Low Persistence
Background process persistence without cleanup guarantee
The skill uses nohup to run devbridge host/connect in the background. These processes may continue running after the skill completes if not explicitly stopped, creating a persistence risk where local services remain exposed.
SKILL.md:160 Document cleanup steps prominently. Consider tracking PIDs and providing explicit cleanup commands in all execution paths including error paths.
Declared capability vs actual capability
Filesystem Block
Declared NONE
→ Inferred WRITE
SKILL.md:87 — curl|bash writes to ~/.huawei/bin/ and ~/.huawei/devbridge/; not declared Network Block
Declared NONE
→ Inferred READ
SKILL.md:87,139 — downloads CLI from res-hd.hc-cdn.cn; Huawei Cloud relay communication; not declared Shell Block
Declared NONE
→ Inferred WRITE
scripts/devbridge_cmd.sh — eval-based command execution with retry logic; no declaration Environment Block
Declared NONE
→ Inferred READ
SKILL.md:130 — reads PATH for ~/.huawei/bin; devbridge auth reads AK/SK env vars Skill Invoke Block
Declared NONE
→ Inferred READ
SKILL.md:44 — skill_invoke used to chain with other DevBridge commands Suspicious artifacts and egress
Critical Dangerous Command
curl -fsSL https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh | bash SKILL.md:87
Critical Dangerous Command
rm -rf ~ references/cli-installation-guide.md:137
Critical Dangerous Command
curl --proxy <proxy-url> -fsSL <install-url> | bash references/cli-installation-guide.md:161
Medium External URL
https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh SKILL.md:87
Medium External URL
https://noh56s7x-8080.cn-north-4-bridge.myhuaweicloud.com SKILL.md:306
Medium External URL
https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.ps1 references/cli-installation-guide.md:36
Medium External URL
https://devbridge. references/rest-api-reference.md:8
Medium External URL
https://res-hd.hc-cdn.cn references/troubleshooting.md:58
Medium External URL
https://clawhub.ai/user/huaweiclouddev skill-card.md:9
Medium External URL
https://clawhub.ai/huaweiclouddev/skills/huawei-cloud-devbridge-tunnel skill-card.md:47
Dependencies and supply chain
| Package | Version | Source | Known vuln | Notes |
|---|---|---|---|---|
| devbridge-cli | 0.1.12+ (unpinned) | https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh | No | Remote script execution with no version pinning or integrity verification |
| bash | any | system | No | Required shell; eval() usage is a concern |
File composition
11 files · 2755 lines
Markdown 9 files · 2391 linesShell 1 files · 359 linesJSON 1 files · 5 lines
Files of concern · 6
SKILL.md Unsafe curl|bash remote script execution · No allowed-tools declaration in SKILL.md · Unversioned remote download source · Background process persistence without cleanup guarantee · curl -fsSL https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh | bash · https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.sh · https://noh56s7x-8080.cn-north-4-bridge.myhuaweicloud.com
scripts/devbridge_cmd.sh eval() usage in command execution layer
references/troubleshooting.md https://res-hd.hc-cdn.cn
references/rest-api-reference.md https://devbridge.
references/cli-installation-guide.md Dangerous rm command in documentation · rm -rf ~ · curl --proxy <proxy-url> -fsSL <install-url> | bash · https://res-hd.hc-cdn.cn/sharedata/hdspace/devbridge/install.ps1
skill-card.md https://clawhub.ai/user/huaweiclouddev · https://clawhub.ai/huaweiclouddev/skills/huawei-cloud-devbridge-tunnel
Other files · acceptance-criteria.md · cli-command-reference.md · iam-policies.md · verification-method.md · _meta.json
Security positives
The skill implements a legitimate DevBridge tunnel management service with no evidence of credential exfiltration
The adaptation layer (devbridge_cmd.sh) is a well-structured version-agnostic wrapper that handles CLI flag variations
SKILL.md provides comprehensive documentation of all tunnel operations and parameters
skill-card.md documents known risks including credential handling, background processes, and destructive operations
The skill does not use base64 encoding, obfuscation, or hidden functionality
No direct credential theft or data exfiltration patterns detected in the code