The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Jira issue descriptions and GitHub pull request bodies to generate documentation content, creating a surface for indirect prompt injection.
Ingestion points: Data enters the context via scripts/jira_acli.py (which fetches Jira issue fields and comments) and via GitHub PR content (titles, bodies, and diffs fetched using gh or curl).
Boundary markers: The instructions do not specify explicit delimiters or system-level 'ignore' instructions for the external content when processing it, which increases the risk of the model following instructions embedded in the data.
Capability inventory: Across all scripts, the skill has the ability to execute CLI tools (git, acli, gh) via subprocess.run, perform network requests via urllib.request, and write files to the local file system (defaulting to ./rhdh-feature-docs-output/).
Sanitization: No explicit sanitization or validation of the fetched text (e.g., removing Markdown/HTML comments or script-like patterns) is performed before it is interpolated into the workflow.
[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/jira_acli.py, scripts/fetch_rhdh_docs.py, and scripts/setup_acli.py to execute commands for acli (Atlassian CLI), git, and gh (GitHub CLI). While these are legitimate for gathering documentation data and cloning repositories, the skill relies on the proper installation and authentication of these external tools on the host system.
[DATA_EXFILTRATION]: The skill manages a sensitive JIRA_API_TOKEN environment variable used for Jira REST API authentication. It follows security best practices by including explicit 'Token safety' instructions to prevent the token from being printed, logged, or included in generated output, and it uses the token only for authenticated requests to the intended service provider.

rhdh-feature-docs