internal/

Package buildinfo exposes version metadata injected at build time via -ldflags.

Package catalog holds firstrun's human-facing knowledge about common commands: a one-line description of what each does, and whether the command has a native "dry run" mode firstrun can offer instead of executing for real.

Package cli wires the firstrun command-line interface together.

Package config loads firstrun's optional TOML configuration.

Package effect turns a single command's argument vector into a set of predicted, neutral effects — what it reads, writes, deletes, downloads, or executes.

Package llm provides a tiny, dependency-free client for the chat APIs of Anthropic, OpenAI, and Ollama.

Package narrate turns an analyzed plan into a short, plain-English explanation.

Package plan assembles a full, inspectable preview of a command or script: for every step it records what runs, a plain-English summary, the predicted effects, and the risk flags.

Package render formats an analyzed plan (and an optional live syscall trace) for output: a styled terminal view, a plain-text version, Markdown, and stable JSON for scripts.

Package risk applies fast, deterministic heuristics to a command and the effects it is predicted to have, and flags the things a person should look at twice before running: destructive deletions, raw-disk writes, piping a download straight into a shell, privilege escalation, world-writable permissions, and writes into system directories.

Package run executes commands on the user's behalf — but only after firstrun has shown a preview and the user has confirmed.

Package shell parses a shell command or script into a structured, inspectable form using the mvdan.cc/sh parser.

Package trace runs a command under a syscall tracer so firstrun can show you exactly what it does at the system-call level.

Package tui renders firstrun's interactive confirmation menu.