Most URL, secret-name, and backtick detections are documentation-only false positives. However, the skill obeys arbitrary preflight output and follows files outside its audited directory. It also loads hidden home workflows and can trigger plugin installation, creating critical supply-chain risk.
The skill runs onchainos preflight and requires the agent to do exactly what data.action says. It also downloads and runs an installer when missing.
The mandatory text explicitly delegates agent control to command output and bootstraps executable installation. No action allowlist or user approval is required.
- `loggedIn: false` โ user is not logged in. Do **not** query identity. Hand off to the existing wal
The flow loads ../../okx-agentic-wallet/SKILL.md outside the audited skill directory and follows its login instructions. That external content is not covered by this audit.
Each "not registered yet" line on the home invites the user to register that role. If the user repli
The agent loads an external identity registration playbook and follows it to completion. That unaudited dependency can initiate consequential registration actions.
**MUST**: Run the shared preflight **only for the onboarding flow** (ยง1 โ `references/how-to-play.md
The mandatory preflight first reads ../okx-agentic-wallet/_shared/preflight.md outside this skill. That external file can change the instructions executed during onboarding.
The review confirms real boundary risks from sibling-package references and hidden home-directory workflow loading. Most other alerts are false positives from Markdown code spans, official user-facing URLs, and protective credential warnings. No prompt-injection text was found in the reviewed files, but the shared preflight also adds installer and update risk.
The shared preflight can update binaries and skill checkouts, run a forced upgrade command, and download installation scripts when the CLI is missing. A community skill should not trigger tool installation or updates without explicit marketplace-controlled confirmation.
The file directly instructs running preflight and fallback installer behavior, including upgrade and install script steps. The risk is contextual because it depends on external onchainos tooling, but the intent is clear.
Capability review items (17)
These are real local capabilities that may be expected for this skill, so they require review but are not counted as confirmed malicious behavior.
- `loggedIn: false` โ user is not logged in. Do **not** query identity. Hand off to the existing wal
The skill directs the agent to read ../../okx-agentic-wallet/SKILL.md, which escapes the skill package boundary. Loading sibling skill content can import unreviewed instructions.
- If the user came from picking the **Daily brief** option (option `4` in Variant A / option `3` in
The flow auto-loads ~/.onchainos/workflows/daily-brief.md after login. Reading executable prompt content from a hidden home directory is a real boundary and prompt-injection risk.
- If the user came from picking any other **workflow pick** while logged out: automatically load the
The flow auto-loads ~/.onchainos/workflows/<file>.md based on a workflow pick. The dynamic hidden-home path can import local prompt content outside the reviewed package.
- **Status column** โ read the agent's `status` field and map it per [`../../okx-ai/references/ident
The status mapping relies on ../../okx-ai/references/identity-invariants.md outside this skill. Cross-package reads can import instructions that were not reviewed with this package.
Each "not registered yet" line on the home invites the user to register that role. If the user repli
The registration path loads ../../okx-ai/references/identity-register.md outside the skill boundary. That delegates control to external prompt content from a sibling package.
| `1` (User) | `Registering your User identity, hang tight... โณ` | [`../../okx-ai/references/identit
Selecting the User role loads ../../okx-ai/references/identity-register.md outside this package. That is a real package-boundary escape even if the target is intentional.
| `3` (Evaluator) | `Registering your Evaluator identity, hang tight... โณ` | [`../../okx-ai/referenc
Selecting the Evaluator role loads the same ../../okx-ai registration playbook outside this package. This crosses the reviewed skill boundary for a privileged registration flow.
The menu routes a logged-in daily brief pick to ~/.onchainos/workflows/daily-brief.md. Loading hidden-home prompt files is outside the audited skill content.
The alternate menu variant also routes daily brief to ~/.onchainos/workflows/daily-brief.md. The target file is local hidden-home content and is not packaged here.
**MUST**: Run the shared preflight **only for the onboarding flow** (ยง1 โ `references/how-to-play.md
The preflight instruction first tries ../okx-agentic-wallet/_shared/preflight.md, which escapes this skill directory. Cross-skill reads can load unreviewed instructions before the user flow begins.
- If the user came from picking the **Daily brief** option (option `4` in Variant A / option `3` in
The same line instructs the agent to load a hidden-home workflow file. That is legitimate product behavior, but it still reads unreviewed local prompt content.
- If the user came from picking any other **workflow pick** while logged out: automatically load the
The same line instructs the agent to load a hidden-home workflow file selected from a route. The reviewed skill does not contain or validate the target file.
The line instructs loading the hidden daily-brief workflow from the user home directory. This is intentional behavior but still imports unreviewed local content.
The line repeats the hidden daily-brief workflow load for another menu variant. It is a real hidden file read outside the marketplace-reviewed package.