The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface because it reads and processes multiple files from the repository to inform its configuration decisions and to modify instructions for other agents.
Ingestion points: The skill explores the repository by reading .git/config, AGENTS.md, CLAUDE.md, CONTEXT.md, CONTEXT-MAP.md, and package.json, and by executing git remote -v.
Boundary markers: There are no explicit delimiters or instructions to ignore potential malicious content within these files during the exploration and editing phases.
Capability inventory: The skill is capable of executing shell commands through the git, gh, and glab CLIs and performing file write/edit operations on agent-facing documentation.
Sanitization: The skill does not implement specific sanitization or validation logic for the content it ingests from the repository before using it to generate or update files.
[DATA_EXFILTRATION]: The skill accesses local metadata by reading .git/config and running git remote -v. While this is necessary for identifying the upstream hosting provider, it involves the exposure of repository configuration data to the agent context.
[COMMAND_EXECUTION]: The skill provides structured guidance for the agent to utilize the gh (GitHub) and glab (GitLab) command-line tools. This includes patterns for creating, listing, and modifying issues and pull requests, as well as making direct API calls.

setup-matt-pocock-skills