The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes Git commands and coordinates between bundled Python scripts using the subprocess module. Scripts like pipeline_state.py, isolation.py, and pipeline_git.py use subprocess.run to manage the repository's branches, worktrees, and history. scripts/_common.py includes a resolution utility that sanitizes arguments for Windows to mitigate command injection risks when executing .cmd or .bat files.
[DYNAMIC_EXECUTION]: The skill bundle includes components that perform dynamic loading or execution. scripts/path_config.py utilizes importlib.import_module to load internal siblings. scripts/loop_run.py is capable of executing arbitrary shell commands provided in a configuration file (LOOP.md).
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection by processing external project data.
Ingestion points: The skill dispatches a subagent to read and process CHARTER.md, SYNTHESIS.md, ROADMAP.md, and evidence files listed in RESEARCH.md (as seen in SKILL.md).
Boundary markers: While references/operating-rules.md provides operational constraints, the roadmapper brief does not explicitly define delimiters or instructions to ignore embedded prompts in the consumed charter files.
Capability inventory: The skill possesses extensive capabilities including Git branch/worktree manipulation, file system writes to the .project/ directory, and subprocess execution across multiple scripts.
Sanitization: The bundled scripts perform structural and semantic validation of the input files, which helps mitigate but does not entirely eliminate the risk of instructions embedded in the prose sections of the documents.

gsd-path-roadmap