# Yolo Strategy Auto-pilot interview: the agent plays both sides, always picking the recommended option. Same workflow as [interview.md](./interview.md) (including the sizing gate) but the agent answers its own questions via "Johnbot." **v0.20 note (s3-migrate-yolo / #1166):** Yolo now emits only the canonical v0.20 shape (date-prefixed story vBRIEFs in proposed/, full PROJECT-DEFINITION.xbrief.json via task project:render, seeded lifecycle folders, no legacy specification.vbrief.json). See the dedicated ## v0.20 Output Shape section. Legend (from RFC2119): !=MUST, ~=SHOULD, ≉=SHOULD NOT, ⊗=MUST NOT, ?=MAY. **⚠️ See also**: [strategies/interview.md](./interview.md) | [strategies/discuss.md](./discuss.md) | [core/glossary.md](../glossary.md) | [vbrief/vbrief.md](../vbrief/vbrief.md) | [artifact-guards.md](./artifact-guards.md) --- ## When to Use - ~ Quick prototyping where speed matters more than precision - ~ When the user trusts the agent's recommended defaults - ? When exploring an idea before committing to a full interview - ⊗ Production systems or compliance-heavy projects — use [interview.md](./interview.md) instead ## Chaining Gate Johnbot auto-selects at the [chaining gate](./interview.md#chaining-gate) too. - ! Johnbot MUST select **"Proceed to specification"** (option 1) — no preparatory detours - ! Johnbot MUST select **"Accept"** at the [acceptance gate](./interview.md#acceptance-gate) — no revisions - ⊗ Ask the real user to choose at either gate — Johnbot handles both automatically ## Sizing Gate Johnbot picks the size too. The same sizing signals from [interview.md](./interview.md#sizing-gate) apply, but Johnbot makes the call without asking the user. - ! Check `PROJECT.md` for `**Process**: Light` or `**Process**: Full` — if declared, use that path - ! If not declared, Johnbot picks based on feature count and complexity signals - ~ Default to Light for typical yolo projects (speed over ceremony) ## Workflow Overview ```mermaid flowchart LR subgraph yolo ["Yolo Strategy"] G{"⚖️ Sizing Gate
Johnbot picks"} I_L["💬 Auto-Interview
Light path"] I_F["💬 Auto-Interview
Full path"] P["📄 PRD
Auto-approved"] S["📁 xbrief/proposed/ (YYYY-MM-DD-*.xbrief.json)
📋 PROJECT-DEFINITION.xbrief.json
v0.20 shape"] end G -->|"Light"| I_L G -->|"Full"| I_F I_L -->|"Johnbot picks defaults"| S I_F -->|"Johnbot picks defaults"| P P -->|"Auto-approved"| S S -->|"Ready"| IMPL["🔨 Implementation"] style G fill:#f0abfc,stroke:#a21caf,color:#000 style I_L fill:#c4b5fd,stroke:#7c3aed,color:#000 style I_F fill:#c4b5fd,stroke:#7c3aed,color:#000 style P fill:#fef08a,stroke:#ca8a04,color:#000 style S fill:#6ee7b7,stroke:#059669,color:#000 style IMPL fill:#7dd3fc,stroke:#0284c7,color:#000 ``` (See ## v0.20 Output Shape for exact artifact rules and the mandatory `task project:render` call.) --- ## Interview Rules Same as [interview.md](./interview.md#interview-rules-shared-by-both-paths), with Johnbot additions: - ~ Use Claude AskInterviewQuestion when available (emulate if not) - ! Ask **ONE** focused, non-trivial question per step - ⊗ Ask multiple questions at once or sneak in "also" questions - ~ Provide numbered answer options when appropriate - ! Include "other" option for custom/unknown responses - ! Indicate which option is RECOMMENDED - ! Pretend you are the user "Johnbot" too - ~ Johnbot asks for details/clarifications on the questions when appropriate - ! Johnbot ultimately goes with the RECOMMENDED option - ⊗ Ask the real user to answer a question — keep working with Johnbot until you can build the specification --- ## Light Path Same as [interview.md Light path](./interview.md#light-path-smallmedium-projects) but Johnbot answers all questions and auto-approves. --- ## Full Path Same as [interview.md Full path](./interview.md#full-path-largecomplex-projects) but Johnbot answers all questions and auto-approves the PRD. --- ## v0.20 Output Shape (s3-migrate-yolo / #1166) This strategy has been migrated to the full v0.20 output shape so yolo-generated projects are accepted by the build skill Pre-Cutover Detection Guard with zero errors on first attempt (resolves the yolo row from the #1166 inconsistency table). - ! Seed the five lifecycle folders under `xbrief/` if any are missing: `proposed/`, `pending/`, `active/`, `completed/`, `cancelled/`. - ! Emit story scope items exclusively as date-prefixed scope vBRIEFs: `xbrief/proposed/YYYY-MM-DD-.xbrief.json` (use the run's creation date for the prefix; choose descriptive slugs). Decompose the yolo plan into one or more focused, buildable story vBRIEFs (v0.6 schema) rather than a monolithic legacy spec. - ! After the proposed/ stories are written, invoke `task project:render` (run from the repo root) to generate/refresh the complete `xbrief/PROJECT-DEFINITION.xbrief.json` (items registry is derived from the lifecycle folders). - ⊗ Never emit `vbrief/specification.vbrief.json` (or any legacy dual-write). Live identity is `xbrief/PROJECT-DEFINITION.xbrief.json`; legacy `vbrief/PROJECT-DEFINITION.vbrief.json` is read-accepted until `deft migrate:xbrief`. - ~ `SPECIFICATION.md` at the project root, if produced at all, must be only a read-only derivative (e.g. via `task spec:render` after the vbriefs exist) that includes the v0.20 deprecated-redirect sentinel. The source of truth is the xbrief/ lifecycle stories + PROJECT-DEFINITION. - ! Before writing any proposed/ stories or PROJECT-DEFINITION, follow the guards in [artifact-guards.md](./artifact-guards.md) (Preparatory Guard for scope items in proposed/; Spec-Generating Guard for PROJECT-DEFINITION). - ! After emitting the proposed/ story vBRIEF(s), surface the GitHub-issue tracking hint from [emit-hints.md](./emit-hints.md) — name all three patterns (none / `--umbrella` / `--per-vbrief`). - ! Final output tree must pass the deterministic v0.20 strategy output validation gate (s2-deterministic-gate) and the build Pre-Cutover Detection Guard with zero warnings/errors. See full acceptance in the yolo migration vBRIEF and the 1166 decomposition. - ~ Once available, cite the canonical contract `strategies/v0-20-contract.md` (s1-contract) for the exact shape. --- ## SPECIFICATION Guidelines Yolo expresses "specification" work via the v0.20 date-prefixed story vBRIEFs emitted to `xbrief/proposed/` (see v0.20 Output Shape section above). The detailed guidelines from [interview.md](./interview.md#specification-guidelines-both-paths) for content quality, requirements IDs, phasing, parallelism, test-first, task sizing, and format still apply — but the *delivery mechanism* is the discrete vBRIEF stories + PROJECT-DEFINITION (never the legacy specification.vbrief.json). --- ## Artifacts Summary (v0.20) **Light path and Full path (identical under yolo; PRD auto-approved on Full):** | Artifact | Purpose | Created By | |----------|---------|------------| | `xbrief/proposed/YYYY-MM-DD-*.xbrief.json` (one or more) | Focused story scope items (date-prefixed per vbrief convention) | Yolo (Johnbot) | | `xbrief/PROJECT-DEFINITION.xbrief.json` | Project identity gestalt + complete scope items registry | `task project:render` (invoked by Yolo) | | `xbrief/{proposed,pending,active,completed,cancelled}/` | All five lifecycle folders seeded | Yolo | | (optional derivative) `SPECIFICATION.md` | Human-readable plan (includes deprecated-redirect sentinel) | `task spec:render` (if invoked) | **Pre-v0.20 / legacy artifacts that MUST NOT be produced by this strategy:** - `vbrief/specification.vbrief.json` - Primary handoff `SPECIFICATION.md` at project root --- ## Invoking This Strategy ``` /deft:run:yolo [project name] ``` Or explicitly: ``` Use the yolo strategy to plan [project]. ``` After completion (v0.20 shape): ``` task project:render # Review date-prefixed stories in xbrief/proposed/ + the generated PROJECT-DEFINITION.xbrief.json # Implement per the v0.20 artifacts (build skill accepts cleanly on first try) ```