The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from repository files and history to generate Pull Request titles and bodies, which could potentially be manipulated by an attacker who has committed code to the repository.
Ingestion points: Processes PR templates from standard locations (e.g., .github/pull_request_template.md), domain context files (CONTEXT.md), and Git history for merged PR titles.
Boundary markers: Instructions explicitly direct the agent to preserve template structure, checklists, and HTML comments, and to quote UI labels verbatim rather than paraphrasing.
Capability inventory: The skill possesses capabilities for PR creation, branch reconstruction via Git, and GitHub CLI operations.
Sanitization: It mandates verification of user-visible strings against repository evidence and requires the agent to surface any mismatches between external data and the repository source to the user.
[COMMAND_EXECUTION]: The skill utilizes local shell commands for Git operations and the GitHub CLI extension to manage branch reconstruction and PR submission.
Evidence: Executes commands including git restore, git commit, git rev-parse, and gh stack submit.
Context: These operations are restricted to specific, verified tool chains and are gated by a mandatory "Publication boundary" checkpoint that requires current-turn user approval before any remote mutation.
[EXTERNAL_DOWNLOADS]: The skill depends on the official github/gh-stack GitHub CLI extension for managing stacked Pull Requests.
Source Identification: The instructions require the extension to be pre-installed and explicitly verify its provenance from the official repository (github/gh-stack).
Context: Use of this well-known service extension is restricted to its official version and includes instructions for verifying the tool's behavior via help commands before execution.

tk-pr-open