v-copy-me
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the current conversation, transcripts, or user-supplied file paths to infer and codify workflow patterns.
- Ingestion points: Step 0 in
SKILL.mdidentifies the source as the current conversation, pasted transcripts, or file paths provided by the user. - Boundary markers: The skill defines non-negotiable rules (Rule 5) instructing the agent to only codify patterns actually observed and to stop if input is insufficient.
- Capability inventory: The skill can write files to the local filesystem, perform Git operations (add, commit, push) to a public repository, and execute
npxcommands to install packages. - Sanitization: Step 4 implements a mandatory sanitization gate that scans for and removes sensitive information such as API keys, email addresses, and absolute file paths before publishing.
- [EXTERNAL_DOWNLOADS]: The skill utilizes the
npxtool to fetch and execute theskillspackage. It also facilitates the download and installation of skills from the author's public GitHub repository (MamounHisham1/vstack) to the user's local environment. - [COMMAND_EXECUTION]: Multiple shell commands are executed to manage the skill lifecycle, including
git(status, add, commit, push),gh(repo view), andnpx(skills add/list). - [DYNAMIC_EXECUTION]: The skill dynamically generates new instruction sets (
SKILL.mdand.opencode/commands/<id>.md) based on runtime session data. These generated files are then installed and loaded as functional agent extensions, which can lead to the propagation of malicious instructions if the source data is compromised.
Audit Metadata