The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the installation of external agent skills using the command npx skills add <owner>/<repo> -y. This enables the execution of unverified code from arbitrary GitHub repositories within the agent's environment.
[DATA_EXFILTRATION]: During the tool discovery phase, the skill extracts package names and technology aliases from manifest files (like package.json or requirements.txt) and sends them to external web search engines and registries. This process effectively leaks the project's internal technology stack to third-party services.
[COMMAND_EXECUTION]: The skill utilizes several shell commands to perform its tasks, including git for repository analysis and npx for managing and installing external tools.
[EXTERNAL_DOWNLOADS]: The skill performs automated web searches to identify relevant tools and downloads code directly from GitHub repositories during the skill installation process.
[PROMPT_INJECTION]: The skill has an inherent exposure to indirect prompt injection because it must ingest and interpret untrusted data from the repository to determine feature completion and document status.
Ingestion points: The skill reads repository manifest files, source code, and documentation files located in the docs/ directory.
Boundary markers: The skill employs OUTPUT-STYLE and ROOT-FIELD-CONTRACT markers and provides explicit instructions for 'surgical' line-based edits to prevent the agent from being manipulated into rewriting entire files.
Capability inventory: The skill possesses the ability to execute git and npx commands, perform web searches, and has broad permissions to read and write files across the repository.
Sanitization: There is no evidence of explicit sanitization or filtering of the content read from the repository before it is processed by the agent to make synchronization decisions.

sync