The Agent Skills Directory

[PERSISTENCE]: The skill attempts to persist its execution path by appending a shell export command to the environment configuration file identified by the CLAUDE_ENV_FILE variable. This modification ensures the tools are available in the system PATH for subsequent sessions. Evidence: printf "export PATH='%s':\"\$PATH\"\n" "${GSD_TOOLS%/*}" >> "$CLAUDE_ENV_FILE".
[COMMAND_EXECUTION]: The skill executes shell scripts that perform extensive scanning of the user's home directory. It checks for the existence of gsd-core/bin/gsd-tools.cjs across numerous hidden configuration directories belonging to different AI agents and development environments, including .claude, .cursor, .windsurf, .trae, and others.
[DYNAMIC_EXECUTION]: The skill implements a runtime discovery mechanism to locate and run a JavaScript utility. It defines a shell function gsd_run that executes the discovered gsd-tools.cjs file using node based on paths computed during the execution of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill reads and displays content from PLAN.md and SUMMARY.md files stored in the .planning/quick/ directories.
Ingestion points: PLAN.md (description line) and SUMMARY.md (status and last action fields) are read via shell commands in SKILL.md.
Boundary markers: No specific DATA_START/DATA_END markers are used when displaying the extracted file content to the user or agent context.
Capability inventory: The skill possesses Bash, Agent, Write, and Edit capabilities, which could be influenced by malicious content embedded in the ingested plan files.
Sanitization: The skill mitigates this risk by strictly sanitizing user-provided slugs ([a-z0-9-]) and stripping non-printable characters or path separators from directory names before display.

gsd-quick