CLI reference
Seven commands. Full flag detail for each is in reelier <command> --help.
| Command | What it does |
|---|---|
reelier init [--yes] | The guided 60-second first run: detect an MCP config, record (demo or real), compile, replay, print a receipt. |
reelier mcp --wrap "..." | Starts the recording proxy on stdio: re-exposes each wrapped downstream MCP server’s tools 1:1, plus 3 reelier_* control tools. |
reelier trace <file>.jsonl | Pretty-prints a trace file. |
reelier compile <file>.jsonl [-o out.skill.md] [--force] | Compiles a trace into a runner-ready SKILL.md, deterministically — zero LLM calls. Refuses to overwrite an existing file without --force. |
reelier run <skill>.skill.md [flags] | Replays a skill. --dry-run prints filled actions without executing. --var name=value passes input variables. --yes allows destructive steps. --wrap replays against live MCP downstream(s). --max-level 1|2 opts into self-heal (BYOK flags apply only on divergence). |
reelier bench <skill>.skill.md | Summarizes a skill’s run-record history: per-level step counts, escalation attempts vs. heals, total LLM tokens. |
reelier push <skill>.skill.md [--all] [--dry-run] [--with-skill] | Syncs new run records (and, on first push, the skill file) to Reelier Cloud. Fully opt-in — does nothing unless REELIER_CLOUD_URL and REELIER_CLOUD_KEY are set. |
Exit codes
reelier run exits 0 if every step passed or was unchecked, 1 on any failure. reelier push exits non-zero only when the batch stopped early on a transient outcome (auth failure or network error) — not merely because some records were permanently rejected.