The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from repository-hosted files such as pull request templates, contribution conventions, and existing pull request descriptions. * Ingestion points: Processes content as described in SKILL.md and references/pr-description-writing.md, which includes reading local project templates and remote pull request metadata. * Capability inventory: Execution of git and gh commands for staging, committing, pushing, and creating/updating pull requests, involving both local file system writes and network operations to GitHub. * Boundary markers: The instructions provide guidelines for scoping (staging explicit files) and auditing results, but they lack specific delimiters or warnings to ignore instructions that might be embedded within the processed external templates or descriptions. * Sanitization: No explicit sanitization, escaping, or filtering of the ingested template content is specified before it is used to generate pull request bodies.
[COMMAND_EXECUTION]: The skill relies on shell commands to interact with the version control system and the GitHub CLI. * Evidence: The workflow involves operations such as fetching remote branches, committing scoped changes, pushing branches, and updating pull request descriptions via temporary files, all of which imply the use of git and gh commands.

open-pr