The Agent Skills Directory

[COMMAND_EXECUTION]: The skill leverages standard git, gh, and jq commands to collect process-related metadata such as PR counts, cycle times, and commit frequencies. These operations are limited to data extraction for delivery performance metrics and do not involve system-level configuration changes or unauthorized privilege escalation.
[SAFE]: The skill analyzes repository metadata to compute team-level signals. The instructions contain explicit guardrails against individual productivity surveillance and focus strictly on aggregate process health (e.g., bus factor and rework rates). No access to sensitive files like credentials, environment variables, or private SSH keys is requested or attempted.
[PROMPT_INJECTION]:
Ingestion points: The skill reads untrusted data from the repository being analyzed, specifically through gh pr list and git log (e.g., Pull Request titles, author names, and commit messages).
Boundary markers: The skill relies on structured report templates (scorecards and graded reports) to organize its output, though it does not explicitly use delimiters for untrusted text in the ingestion phase.
Capability inventory: The skill is limited to local data processing and reporting; it does not possess capabilities to perform network exfiltration to unknown domains or write to sensitive system files.
Sanitization: Uses jq for all data parsing and transformation, ensuring that repository strings are treated as literal data rather than executable instructions during the measurement phase.

delivery-health