The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs several shell operations and executes local scripts. It uses git rev-parse HEAD to determine repository state and frequently invokes Python scripts from a hidden local directory, such as python3 .codestable/tools/codestable-goal-consistency-gate.py. These scripts are not provided within the skill package, meaning the skill's behavior depends on the integrity of the files in the target project's .codestable directory.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes multiple external data sources to generate its execution plans and instructions.
Ingestion points: The skill ingests data from roadmap documents ({slug}-roadmap.md), item lists ({slug}-items.yaml), and feature design files ({feature-slug}-design.md) found within the user's project repository.
Boundary markers: There are no specific instructions or delimiters provided to the agent to distinguish between its own operational guidelines and the content within these processed files.
Capability inventory: The skill possesses significant capabilities, including executing shell commands, writing files to the project directory, and generating complex, multi-step slash commands for the user to execute.
Sanitization: The skill lacks explicit validation or sanitization of the input file content before using it to construct instructions or determine the execution flow.

cs-roadmap-impl-goal