Back to skill

Security audit

sn-proactive-agent

Security checks for vulnerabilities and agentic risk

Overview

This skill is transparent about its goal, but installing it would place an unreviewed external runtime in a high-trust position over Hermes conversations and session actions.

Install only if you trust the publisher and the exact release artifact. Before setup, confirm the v0.1.3 release is final, verify the wheel, understand that complete Hermes Q&A content will be stored locally and processed by the service, review the Hermes files it will modify in dry-run mode, and keep a rollback path for the Hermes source/configuration and user data directory.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T08 · Insecure Dependencies

Warning
Location
references/install/overview.md:14
Finding
Unverifiable External Package Receives High-Trust Hermes Integration## Vulnerability Details **File Location**: `references/install/overview.md:14-25, 46-63`; `references/connectors/hermes.md:52-60, 90-96` **Vulnerability Type**: T08: Insecure Dependencies **Risk Level**: Medium The Skill instructs users to download and install an external Python wheel that is not included in the audited project. The documentation states that version `0.1.3` has not yet been published under the new name, so neither its source nor its final artifact could be reviewed during this audit. ### Relevant Code Segments `references/install/overview.md:46-63`: ```text gh release view v0.1.3 --repo OpenSenseNova/SenseNova-Skills-ProactiveAgent --json tagName,isDraft,isPrerelease,assets gh release download v0.1.3 --repo OpenSenseNova/SenseNova-Skills-ProactiveAgent --pattern sn_proactive_agent-0.1.3-py3-none-any.whl --pattern SHA256SUMS ``` ```text pipx install --python "$SNPA_PYTHON" "$SNPA_PACKAGE" ``` `references/connectors/hermes.md:52-60`: ```text sn-proactive-agent setup --harness hermes --web-only --hermes-root "$SNPA_HERMES_ROOT" --dry-run ``` ```text sn-proactive-agent setup --harness hermes --web-only --hermes-root "$SNPA_HERMES_ROOT" ``` The associated instructions state that the installer backs up and changes the Hermes TUI entry point and build artifacts, adds and builds a Web-only bridge, installs hooks and Python observation components, and enables a plugin. `references/connectors/hermes.md:90-96` documents that the resulting bridge uses the current window's native `prompt.submit` interface to submit approved actions and that completed turns contain the full question-and-answer content. ### Technical Analysis The effective runtime and connector implementation are supplied by a future private GitHub Release rather than by files present in the audited project. Consequently, the implementation that receives access to conversation data and modifies Hermes cannot be verified from this rep ...[truncated 2708 chars]
Remediation
## Remediation Suggestions 1. Include the exact runtime and connector source corresponding to the distributed wheel in an auditable tagged release. 2. Publish reproducible build instructions, locked dependencies, and build provenance so reviewers can verify that the wheel corresponds to the reviewed source. 3. Sign the release artifact using a pinned, independently trusted maintainer identity, such as Sigstore provenance or another verifiable code-signing mechanism. 4. Pin and document the expected artifact digest in a separately authenticated channel instead of relying only on a checksum distributed beside the wheel. 5. Require verification of both the cryptographic signature and the exact version before installation. 6. Document every file, configuration key, hook, plugin, build artifact, endpoint, and data field accessed or modified by the connector. 7. Obtain explicit user consent before enabling full question-and-answer capture, with a lower-data or redacted mode where possible. 8. Minimize connector privileges and filesystem access. Run the service under a dedicated account or operating-system sandbox where practical. 9. Restrict the service to loopback interfaces by default and apply authentication if access beyond loopback is supported. 10. Add an auditable uninstall and rollback verification process that confirms restoration of modified Hermes files and removal of connector configuration. 11. Do not enable installation instructions for version `0.1.3` until the final artifact, source tag, provenance, and signatures are available and have been independently reviewed.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Findings (3)

Ae1

High
Category
analysis-evasion
Content
- 本 Skill 的目录和文件名使用英文;除标准入口 `SKILL.md` 外,使用小写与连字符。
Confidence
100% confidence
Finding
Referenced artifact was not completely inspected

Natural-Language Policy Violations

Medium
Confidence
94% confidence
Finding
The natural-language content in the skill manifest and operational instructions is exclusively Chinese, and line 35 further standardizes explanatory text in Chinese. This creates a language-policy issue because the skill does not offer users a language choice or document that the Chinese-only constraint is required for a justified region-specific purpose.

Natural-Language Policy Violations

Medium
Confidence
95% confidence
Finding
This file forces a specific language for all user-facing instructions, which can violate language or locale policy when no user opt-in or justified regional constraint is provided. The document does not state that it is intended only for Chinese-speaking users or offer an alternative language path.

Static analysis

No suspicious patterns detected.