# Invocation model

The toolkit is agent-client agnostic. This taxonomy guides people and agents. It is an
**invocation policy**, not a runtime: the CLI never authors SDD artifacts and never runs a
skill loop.

The **canonical workflow path** is Plan → Prompt → Implement → Check → PR → Review → Fix →
Validate. Invoke `saf-route` when the next step is uncertain. Read the
recommended skill's local `SKILL.md` as the source of truth.

Missing project configuration resolves to `apply + supervised`. `config policy`
can persist an explicit policy override. Autonomous does not mean unlimited
authority. When policy is `full + autonomous`, the **invoking agent** may read
the next on-path `SKILL.md` for a normal or authorized repair transition when
the guardrails admit that transition. A recoverable negative result may route to its owning
repair Skill; exceptional authority, safety, budget, no-progress, or human-override conditions
return control to a human. The CLI still does not invoke Skills. Commit, push, merge, tag, and
publish stay human on every preset.

## User-invoked orchestrators

- `saf-route`
- `saf-brainstorm`
- `saf-create-spec`
- `saf-explain`
- `saf-create-prompts`
- `saf-implement`
- `saf-implement-multi`
- `saf-create-pr`
- `saf-validate`

## Agent-assisted review discipline

- `saf-check-task`
- `saf-review-pr`
- `saf-fix-pr`

Invoke `saf-route` when the next step is uncertain. Read the recommended skill's local `SKILL.md` as the source of truth. See the [skills catalog](skills-catalog.md) for all 12 public skills.