The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands including pwd, git status, and git log --oneline -5 to determine the current working directory and synchronize project state with version control history.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection by reading and processing content from local project files into the agent's active context.
Ingestion points: The agent reads contents from PROJECT_CONTEXT.md, CLAUDE.md, package.json, pyproject.toml, and requirements.txt within the current project directory.
Boundary markers: There are no explicit delimiters or specific instructions to ignore embedded prompts when these files are read and processed.
Capability inventory: The skill maintains the ability to execute shell commands (via git) and perform file system read/write operations based on the state derived from these files.
Sanitization: The content retrieved from the local environment is used directly without sanitization or validation to update the session's focus items and project status.

project-context