The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and analyze user-generated content from the drupal.org issue queue and RSS feeds (e.g., using WebFetch or curl). Because this content is user-contributed, it represents a potential surface for indirect prompt injection where malicious instructions could be embedded in issue descriptions or comments to influence agent behavior.
Ingestion points: SKILL.md (instructions to fetch issue details), references/issue-queue-rss-feeds.md (examples of RSS feed ingestion).
Boundary markers: The skill does not provide delimiters or warnings to treat fetched issue content as untrusted.
Capability inventory: The skill documents extensive use of shell commands and project configuration tools (composer, drush, git) that can be influenced by the content of fetched patches or issue metadata.
Sanitization: No sanitization or validation of the remote content is suggested before processing.
[COMMAND_EXECUTION]: The skill provides numerous instructions and shell script examples for executing commands related to project maintenance, including module updates, database updates, and patch application. This behavior is expected and consistent with the skill's primary purpose.
Evidence: Multiple shell script examples in the examples/ directory and command snippets in SKILL.md utilizing composer require, drush updb, and git apply.
[EXTERNAL_DOWNLOADS]: The skill directs the user and the agent to download external resources, primarily Drupal patches, from drupal.org and git.drupalcode.org.
Evidence: Patch URLs in SKILL.md and examples/find-and-apply-patch.sh referencing https://www.drupal.org/files/issues/ and https://git.drupalcode.org/.
Note: These domains are official infrastructure for the Drupal ecosystem and are recognized as legitimate services for the intended use case.

drupal-contrib-mgmt