The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitHub, which serves as an ingestion point for potential indirect prompt injection.\n
Ingestion points: Content is retrieved from pull request descriptions, diffs, review threads, and linked issues through the GitHub CLI as defined in the gather step of SKILL.md.\n
Boundary markers: Instructions explicitly command the agent to treat forge-derived text as untrusted inputs to be graded, rather than instructions to be followed or tool-use overrides.\n
Capability inventory: The skill can execute GitHub actions like merging pull requests via gh pr merge and managing local temporary files for data processing.\n
Sanitization: The skill uses structured GraphQL queries with variables and the jq tool to parse data, reducing the risk of injection through malformed inputs.\n- [COMMAND_EXECUTION]: The skill utilizes the GitHub CLI (gh) and a local bash script to perform its operations.\n
It executes various gh commands (pr view, issue view, api graphql) to collect data for the review.\n
It executes gh pr merge to perform the merge action after user confirmation according to references/merge-execution.md.\n
The included scripts/fetch-pr-history.sh script is used to automate data retrieval and includes basic validation for input parameters such as the repository name and PR number.

checking-merge-readiness