`reyn run`¶

Run a skill end-to-end.

Synopsis¶

reyn run [OPTIONS] [SKILL] [INPUT]

Name	Description
`SKILL`	Skill name. Resolved in order: `reyn/project/<name>` → `reyn/local/<name>` → `src/reyn/stdlib/skills/<name>`.
`INPUT`	Initial input. JSON string is used as-is (must be a valid artifact). Natural-language string is auto-wrapped as `{"type": "user_message", "data": {"text": "..."}}`. Reads stdin when omitted.

Flag	Description
`--skill-path DIR`	Path to a skill directory (overrides name resolution).
`--module MODULE`	Python module path exposing a `skill` object.
`--skill-root DIR`	Root of the skill tree for shared artifact/phase resolution. Inferred automatically when using `--skill-path`; override when the inferred root is wrong.
`--model MODEL`	Model class (`light` / `standard` / `strong`) or LiteLLM model string. Resolved via `reyn.yaml`'s `models` map.
`--output-language LANG`	Output language code. Default from `reyn.yaml`.
`--max-phase-visits N`	Cap on single-phase revisits per run. `0` = unlimited. Default `25`.
`--events`	Print the full event log after execution.
`--strict`	Enforce required fields at every nesting depth (default: top-level only).
`--allow-shell`	Enable the `shell` Control IR op. Off by default.
`--allow-unsafe-python`	Enable unsafe-mode Python preprocessor steps (no AST sandbox). `--allow-untrusted-python` is a legacy alias for backwards compatibility.

Run a stdlib skill with natural-language input:

reyn run direct_llm "reyn is a workflow OS for LLMs."

Run with structured JSON input:

reyn run my_skill '{"type": "topic_input", "data": {"topic": "ml"}}'

Run from stdin:

echo "summarize this text" | reyn run direct_llm

Replay events afterwards:

reyn run direct_llm "..." --events

Run a meta-skill that needs shell access:

reyn run skill_improver "improve my_skill" --allow-shell