The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests data from local project files, such as .codestable/lessons/ and attention.md, to inform its implementation strategy. This represents an indirect prompt injection surface where instructions in the repository could influence agent behavior. The skill mitigates this by instructing the agent to verify "lessons" against canonical code and documentation before adoption.
Ingestion points: Reads files from the .codestable/ directory, including lessons, requirements, and attention notes.
Boundary markers: The agent is directed to perform "read-repair" and verify facts before applying lessons.
Capability inventory: File system writes (.codestable/work/feat-{slug}.md), git operations, and sub-agent instantiation.
Sanitization: Standard for local file ingestion in developer tools.
[COMMAND_EXECUTION]: The skill uses standard version control commands, including git diff for staged changes and git commit for milestone creation. It also invokes local CLI tools to manage the creation of sub-agents during the code review process.
[REMOTE_CODE_EXECUTION]: The skill orchestrates the creation of sub-agents and reviewers using local or platform-native CLI tools. It specifies model selection criteria and delegation methods to perform heterogenous code reviews as part of the intended development lifecycle.

cs-feat