STOP - Read before ANY tool call
Read _shared/skill-boilerplate.md first. It holds the guardrails read, the session gate rule, the user sim rule, the one-step rule, the gate check, and the user chat language rule. The steps below are skill-specific only.
Step -1b - Caveman Mode (Full cap)
- Read
{{SDD_ROOT}}/preferences.json(create{ "caveman_mode": false, "caveman_level": "full" }if missing). - If
caveman_modeis false: continue without compression. - If true: load
{{TOOLKIT_ROOT}}/skills/_shared/caveman/CAVEMAN.md; apply Full participation cap + prefscaveman_level(Lite skills never escalate); show once:[Caveman] Modo ativo (respostas compactas, level={effective}). Digite caveman off para desativar. - Honor
caveman on|off|status|lite|full|ultra(andstop caveman/normal mode) during the session. - Auto-Clarity + never-compress gates/drafts/paths per
CAVEMAN.md.
Trigger
Negative signal: applies as defined in _shared/skill-boilerplate.md; the skill-specific signals are listed below.
Use when user asks for /developer or requests generic coding/refactoring tasks without specifying a stack.
Outcome
Correct stack skill loaded and executed, or ad-hoc implementation in fallback mode with optional handoff to /commit.
Lazy-load (only when needed)
| When | Path (after syncing the active adapter) |
|---|---|
| Structure / quality (when coding) | {{TOOLKIT_ROOT}}/skills/_shared/code-guidelines/principles/structure-and-quality.md |
| Git / language policy | {{TOOLKIT_ROOT}}/rules/branch-validation.mdc |
| Developer flow | {{TOOLKIT_ROOT}}/skills/_shared/developer-common/GUIDE.md |
| Subagent-first / SPAWN.md | {{TOOLKIT_ROOT}}/skills/_shared/developer-common/subagent-first.md, {{TOOLKIT_ROOT}}/skills/_shared/agents/SPAWN.md |
| Caveman Mode (if active) | {{TOOLKIT_ROOT}}/skills/_shared/caveman/CAVEMAN.md |
| Reference index (routing only) | {{TOOLKIT_ROOT}}/skills/developer/reference.md |
| Process step detail (lazy) | {{TOOLKIT_ROOT}}/skills/developer/references/<section>.md |
Do not load dev_persona or Antigravity KI artifacts.
Never by default: do not preload all references/*.md or all stack guideline packs. Route first; load one section per Process step (SKILL-REFERENCE-RETRIEVAL.md).
Reference routing
| Situation | Path |
|---|---|
| Frontend / design brief | references/frontend-design-routing.md |
| Stack match table | references/stack-routing.md |
| Fallback execution | references/fallback-execution.md |
| Must not (full) | references/must-not.md |
Process
Read references/<section>.md for procedural detail — not full reference.md.
0. Frontend design routing
Follow references/frontend-design-routing.md when the task touches UI.
1. Stack routing
Follow references/stack-routing.md (e.g. pom.xml / build.gradle → java-developer). On match, load that stack SKILL.md and stop this router.
When the user names a story PLAN step (features/.../PLAN/PLAN_*.md plus a step number): load that step block and the matching REFINE/tasks.md boxes before any code. Do not load the rest of the PRD. After the step, set the step status to COMPLETED in the PLAN and mark those boxes - [x]. Without a story path, stay on the ad-hoc fallback.
2. Fallback only
If no stack match: follow references/fallback-execution.md.
Must not
Enforce the full list in references/must-not.md. Critical: route before implement; no guideline dumps into children; no auto-commit.
Handoff
| Situation | Next |
|---|---|
| Commit | /commit |
| .NET work (explicit) | /dotnet-developer |
| Java work (explicit) | /java-developer |
| Large scope | /sdd-spec -> sdd-plan -> sdd-develop |
| UI design / brief | For net-new UI or a substantial visual redesign, create or update docs/DESIGN-BRIEF.md; see frontend-guidelines/frontend-practices.md. |
| New Blip plugin scaffold | /blip-plugin-developer |