The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the GitHub CLI (gh) to perform project discovery, authentication checks, and GraphQL queries. This is the primary mechanism for the skill's operation.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from GitHub (project field names, view names, and item titles) and writes it into .ghpm/config.json. While this presents an attack surface where an attacker could place malicious instructions in GitHub metadata, the risk is inherent to the skill's purpose. 1. Ingestion points: Fetches project metadata via gh project field-list, gh api graphql, and gh project item-list. 2. Boundary markers: None; data is written to structured JSON files. 3. Capability inventory: File system write access and execution of gh commands. 4. Sanitization: The skill transforms layout names but does not sanitize other metadata fields.
[PERSISTENCE]: Offers to modify .claude/settings.local.json to install session hooks. This persistent change to the agent environment is disclosed in the documentation and requires user confirmation.

ghpm-init