v6.4.3 to v6.5.0

---

name: saf-route

description: Recommend the next local SDD skill without changing files. Use when a user needs help choosing a safe workflow step or resolving prerequisites.

metadata:

extends: null

requires: [config]

consumes: [discovery-state]

produces: [route-recommendation]

baseline: []

depends_on: []

conflicts: []

requires_cli: null

autonomy_profile:

supported_levels: [manual, supervised, autonomous]

auto_continue_condition: 'route recommendation present and one on-path next Skill is admissible; this Skill remains read-only'

blocking_conditions: [ambiguous_state]

evidence_required: [route-recommendation]

---

# Route an SDD workflow

## When to use

Use before a workflow step when the requested phase, prerequisites, or installed pack are unclear. Recommend the canonical workflow path (Plan → Prompt → Implement → Check → PR → Review → Fix → Validate). Route first by artifact state, then by uncertainty. Read [spec lifecycle](../sdd-agentic-flow-shared/references/spec-lifecycle.md): listing slugs ≠ loading bodies; 0 ask / 1 select / >1 human gate.

## When not to use

Do not use to implement, review, create a PR, change files, or replace the candidate skill's instructions.

## Inputs

- The requested outcome and available local SDD artifacts.

- `.sdd-agentic-flow/config.yml`, when present.

- Installed skill directories and relevant candidate `SKILL.md` files.

## Workflow

1. Read `.sdd-agentic-flow/config.yml` when it exists. If it is missing, recommend `saf-setup`.

2. Read `../sdd-agentic-flow-shared/references/workflow-routing.md`, `../sdd-agentic-flow-shared/references/workflow-safety.md`, and `../sdd-agentic-flow-shared/references/spec-lifecycle.md`.

3. Inspect the candidate local `SKILL.md` before stating what it does. Its instructions are the source of truth.

4. Match the request against the routing table in `../sdd-agentic-flow-shared/references/workflow-routing.md` — it is the single source of truth for routing situations and recommended skills; do not reproduce or re-derive the table here, and never let this step's wording diverge from it.

5. Identify missing packs, prerequisites, unresolved uncertainty, and any human decision required. Treat workflow/artifact state as the primary router: use uncertainty only to distinguish conversational brainstorming from durable discovery or a human gate. A discovery-only feature workspace is resumable with `saf-brainstorm`, but is not a spec-ready package. If the request needs a feature package and does not name one, listing directory names under `specs.root` and skimming `context.md` or discovery status is allowed. 0 matches → ask; 1 unique → name it in the recommendation; >1 plausible → human decision required; never “probably this one.” Do not read every `spec.md` to make the recommendation. Listing ≠ loading bodies. This remains a read-only router, not a package registry. If config `autonomy_level` is `autonomous`, the invoking agent may follow one admissible on-path recommendation without a new confirmation; this Skill never invokes the next Skill.

## Safety

- This skill is read-only and never invokes another skill automatically.

- Do not infer requirements, install packs, change files, or perform Git, release, deploy, or publish actions.

- Follow `../sdd-agentic-flow-shared/references/workflow-safety.md` and preserve human authority.

## Output

### Recommended route

- Current phase:

- Recommended skill:

- Why this fits:

- Prerequisites:

- Human decision required:

- Next invocation:

- Safety limits:

- Status:

- Next recommended skill:

- Reason:

### Autonomy

Supports `manual`, `supervised`, and `autonomous` autonomy levels. In autonomous mode, the

invoking agent may follow one admissible recommendation without a new human confirmation, while

this Skill remains read-only and never invokes another Skill itself. See

`../sdd-agentic-flow-shared/references/autonomy-guardrails.md`.