The Agent Skills Directory

[SAFE]: The skill performs legitimate initialization tasks, such as creating the .agent/ directory structure and populating it with documentation templates and Grafana configuration files.
[COMMAND_EXECUTION]: The skill configures a PostToolUse hook in .claude/settings.json which executes a local Python script (monitor-tokens.py) to monitor context usage after tool calls. This is a persistent automation within the agent's environment used for its stated purpose.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting untrusted data from project configuration files to populate documentation templates.
Ingestion points: Data is read from package.json, pyproject.toml, go.mod, Cargo.toml, composer.json, and Gemfile using the project_detector.py script.
Boundary markers: The skill does not implement boundary markers or instructions to ignore instructions embedded in the project files during the template replacement process.
Capability inventory: The skill possesses Write, Bash, Read, and Glob capabilities as defined in the frontmatter.
Sanitization: There is no evidence of input validation or sanitization for the project names or stack descriptions before they are interpolated into the generated files.

nav-init