The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes bash commands to list backlog items, read planning documents, and manage directory structures. It also uses a wrapper function to call a vendor CLI tool for metadata updates and commits.\n- [DYNAMIC_EXECUTION]: The skill performs dynamic discovery of the @opengsd/gsd-core tool by searching multiple platform-specific configuration directories (such as .claude, .cursor, .hermes, and others) and executing the resulting path using node.\n- [PERSISTENCE]: The skill attempts to persist the discovered tool's path by appending an export PATH statement to the CLAUDE_ENV_FILE, which ensures the tool is available in the environment for subsequent agent sessions.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests content from project files to process backlog items, creating a potential surface for indirect instructions.\n
Ingestion points: Reads .planning/ROADMAP.md, CONTEXT.md, and RESEARCH.md via cat and grep.\n
Boundary markers: None identified; file content is used to populate variables like ${DESCRIPTION}.\n
Capability inventory: File system writes, directory deletion, and command execution via node and bash.\n
Sanitization: No specific sanitization or escaping is performed on the data extracted from the local files before it is interpolated into shell commands.

gsd-review-backlog