The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell scripts (lookup.sh, detect-version.sh, fetch-docs.sh) to implement its core functionality, such as project version detection, documentation retrieval, and keyword searching. These scripts interact with the local filesystem and use standard Unix utilities, employing best practices like set -euo pipefail for reliability.
[EXTERNAL_DOWNLOADS]: Through the fetch-docs.sh script, the skill downloads official TYPO3 documentation from GitHub repositories (e.g., TYPO3-Documentation/TYPO3CMS-Reference-Typoscript). These sources are well-known and authoritative within the TYPO3 community, making the downloads legitimate and safe for maintaining up-to-date technical references.
[INDIRECT_PROMPT_INJECTION]: The skill processes local project files (composer.json, composer.lock) and remote documentation files. This ingestion of external data constitutes an attack surface where malicious instructions could be embedded to influence the AI agent's behavior.
Ingestion points: scripts/detect-version.sh reads project-specific configuration files from the local workspace. scripts/fetch-docs.sh retrieves documentation content from external GitHub repositories.
Boundary markers: The skill does not explicitly use delimiters or provide specific instructions to the agent to treat the ingested data as untrusted or to ignore embedded commands.
Capability inventory: The agent can execute the provided shell scripts and use the results to perform tasks like code generation and review, creating a path for potential exploitation if the input data were compromised.
Sanitization: Documentation is converted from RST to Markdown using a provided Python script, but no additional sanitization is performed to filter out potentially malicious instruction-like content from the ingested text.

typo3-typoscript-ref