The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation guides the user to install the gh-stack extension using gh extension install github/gh-stack. This targets a well-known repository under the official GitHub organization.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted data from GitHub pull requests (such as branch names, titles, and check statuses) which could potentially contain malicious instructions.
Ingestion points: External data is fetched via gh api and gh pr view in status.sh and approve.sh.
Boundary markers: The instructions in SKILL.md explicitly require the agent to wrap script output in triple-backtick text blocks to prevent the LLM from interpreting the output as instructions.
Capability inventory: The skill utilizes gh and git CLI tools for repository and PR management across its scripts.
Sanitization: External data is processed and filtered using jq to extract specific fields before being displayed or used in further commands, reducing the risk of injection.

gh-stack