llm-wiki

该 skill 的整体能力与“本地知识库构建/维护”目的大体一致，未见明确凭据窃取、第三方中转或恶意外传行为，因此不像恶意技能。主要风险在于：大量未审计的本地脚本与可选第三方提取器、以及处理不受信任外部内容时仍拥有执行命令和写文件能力。综合判断为 SUSPICIOUS（偏中风险），原因是安装与执行信任不足、提示注入面较大，而不是确认恶意。

No strong indicators of intentional malware (no network exfiltration, credential theft, or backdoor behavior) are present in this module. However, the code has moderate security-abuse surfaces: it can execute an attacker-specified Chromium binary via GRAPH_LARGE_PERF_CHROME_EXECUTABLE and it executes a repository bash build script via spawnSync with a filesystem path argument derived from environment-influenced shape values; additionally, GRAPH_LARGE_PERF_ARTIFACT_DIR can redirect where results are written (potential clobber within process permissions). If run in a trusted environment, risk is low; if environment variables can be attacker-controlled, security risk increases substantially.

No strong evidence of intentional JavaScript malware is visible in the provided fragment (no eval/dynamic execution, no network calls, no credential theft logic). The primary security risk is supply-chain/build tooling exposure: the module executes scripts/build-graph-data.sh via bash, passes kbPath as an argument without validation at this boundary, and forwards the entire process.env to the subprocess while suppressing stdout/stderr. Treat this as medium risk pending verification of how kbPath is sourced/validated upstream and how build-graph-data.sh safely handles its arguments and environment.

This module is primarily a LibreOffice-driven formula recalculation tool, but it also installs (writes) a persistent LibreOffice Basic macro into a user profile directory and then triggers that macro via a soffice script URL. Because the actual RECALCULATE_MACRO payload is not included in the provided fragment (and get_soffice_env is external), the code cannot be confirmed as harmless; the macro could be leveraged for sabotage or arbitrary actions depending on its content.

This wrapper executes `soffice` with environment-variable customization and, when Unix sockets are unavailable, writes C source to `/tmp`, compiles it with `gcc`, and injects the resulting shared library into `soffice` via `LD_PRELOAD`. That LD_PRELOAD + runtime compilation pattern is highly suspicious and can enable arbitrary native code execution inside a trusted third-party application. The provided fragment is truncated (the actual `_SHIM_SOURCE` payload is missing), so the true maliciousness cannot be verified from this module excerpt alone; security should be treated as high-risk pending inspection of the shim code.

BENIGN with medium supply-chain risk. The behavior fits a self-updater skill, but it trusts and executes an unpinned remote install script from GitHub, so compromise of that repo or branch would directly affect the user's local skill installation.

No clear evidence of intentional malware, tracking, credential theft, or network-based exfiltration in this fragment. The primary security weakness is use of zipfile.ZipFile.extractall(output_path) on an untrusted Office ZIP without validating member paths, enabling potential ZIP Slip/path traversal and arbitrary file write outside the chosen output directory. Additional risk could be introduced by the unseen DOCX helper functions, but that behavior is not assessable from this module alone.

This module is a document-processing utility that performs headless LibreOffice execution by installing and running a LibreOffice Basic macro from a runtime-writable profile in /tmp. The wrapper itself shows no network access or credential theft, but it provides a strong code-execution primitive (macro write + macro execution via UNO script). Because the macro payload (ACCEPT_CHANGES_MACRO) is not present in the provided fragment, its behavior cannot be verified; if that payload were malicious or tampered, the impact could be substantial within the LibreOffice execution context. Additionally, timeout handling may misreport success, which can undermine output integrity.