The Agent Skills Directory

[SAFE]: The skill implements legitimate developer workflow automation for handling stacked pull requests.
[COMMAND_EXECUTION]: The skill instructions include shell scripts that utilize git and gh (GitHub CLI). These tools are used appropriately within the context of the skill's purpose to query repository state and PR metadata.
[DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration. Network access is restricted to standard GitHub CLI operations directed at official repository metadata. No sensitive local files (like SSH keys or credentials) are accessed.
[INDIRECT_PROMPT_INJECTION]: The skill features an attack surface for indirect injection as it parses data from PR bodies and comments to determine stack membership.
Ingestion points: Reads content from gh pr view --json body and gh pr view --json comments in SKILL.md.
Boundary markers: None present in the parsing logic.
Capability inventory: Execution of gh and git commands based on parsed identifiers.
Sanitization: The skill employs a secure pattern by using grep -oE '#[0-9]+' to extract only numeric PR identifiers. This strict regex validation prevents an attacker from injecting shell commands or prompt overrides via PR comments or body descriptions.

stack-mode