The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run and Popen across multiple scripts (such as scripts/_common.py, scripts/pipeline_git.py, and scripts/detect_project.py) to execute git and gh (GitHub CLI) commands for version control and integration tasks.\n- [COMMAND_EXECUTION]: scripts/loop_run.py includes a shell execution function that runs commands specified in a local LOOP.md configuration file to automate verification loops.\n- [DYNAMIC_EXECUTION]: scripts/path_config.py uses importlib.import_module to load internal sibling modules, which is a standard pattern for modular script organization within the framework.\n- [DYNAMIC_EXECUTION]: scripts/isolation.py programmatically modifies configuration files for AI tools like Claude, Cursor, and Codex (e.g., .claude/settings.json) to install 'guard' hooks that execute a local Python script upon tool usage.\n- [INDIRECT_PROMPT_INJECTION]: The skill's 'Supplied-spec mode' creates an attack surface for indirect prompt injection by ingesting natural language instructions from user-provided project specifications like PRDs or issue descriptions.\n
Ingestion points: Untrusted data enters the agent context via specification files described in SKILL.md.\n
Boundary markers: The skill does not explicitly use delimiters to wrap external document content.\n
Capability inventory: The skill possesses high-privilege automation capabilities including arbitrary git command execution and filesystem modification across all bundled scripts.\n
Sanitization: The scripts perform extensive structural and regex validation on the framework's own state files but lack specific sanitization for the narrative content of external project documents.

gsd-path-define