The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands to resolve project paths and detect git worktrees during its setup phase. These include git rev-parse, realpath, readlink, pwd, and dirname. These operations are limited to environment discovery and path resolution.
[COMMAND_EXECUTION]: The skill runs a bundled Python script (extract-domain-context.py) to perform a lightweight scan of the project. The script extracts file trees, entry points, and metadata, writing the results to a local .understand-anything/ directory within the project root.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted codebase content (source code snippets, README.md, package.json) and passes it to a subagent for domain analysis.
Ingestion points: Reads all source files (limited by extension), package.json, and README.md within the project root.
Boundary markers: Not explicitly defined in the skill instructions; however, the subagent relies on a prompt template stored in $PLUGIN_ROOT/agents/domain-analyzer.md.
Capability inventory: The skill can execute local shell commands, run a Python script, write to the local filesystem, and dispatch subagents.
Sanitization: The Python script performs truncation and filtering of files based on .gitignore, but does not perform specific sanitization of code content for instructions.

understand-domain