The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill parses and executes scripts from repository configuration files locally.
Evidence: references/ci-checks.md directs the agent to identify CI configuration files (GitHub Actions or GitLab CI) and run run: or script: elements directly in the shell. This provides an execution path for scripts controlled by the repository's contents.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted template files from the repository to generate pull request descriptions.
Ingestion points: SKILL.md and references/pull-request-template.md ingest content from .github/PULL_REQUEST_TEMPLATE.md or .gitlab/merge_request_templates/Default.md.
Boundary markers: Absent. The agent is not provided with explicit delimiters or instructions to ignore malicious directives within these templates.
Capability inventory: The skill uses Bash to execute gh and glab CLI tools for repository management and network interaction.
Sanitization: Absent. The agent incorporates template content into its generated output without filtering or validation.
[COMMAND_EXECUTION]: Use of Git and platform-specific CLI tools for operations.
Evidence: SKILL.md and references/pr-cli.md employ git, gh (GitHub CLI), and glab (GitLab CLI) to perform staging, committing, pushing, and pull request creation.
[EXTERNAL_DOWNLOADS]: Use of utility runners that may fetch code at runtime.
Evidence: references/ci-checks.md utilizes npx to execute tools such as markdownlint-cli2 and cspell, which triggers a download from the NPM registry if the package is not already present in the environment.

create-pr