Loading
Execute the current intention via codegen. Use when acting on state/intention.json against the environment.
Recommended by author
This prompt takes no variables — just pick a model and run.
# Plan Convert the current intention into exactly one concrete environment action for this tick via `codegen`. This skill is a tick-level action controller. It does not complete the whole goal at once. After one meaningful action, stop. ## Inputs | File | Use | | ----------------------- | ----------------------------------------------- | | `state/intention.json` | Current goal or intention | | `state/observation.txt` | Available actions, context, affordances | | `state/needs.json` | Optional authoritative need levels | | `state/emotion.json` | Optional `needs` snapshot if needs.json missing | | `state/plan_state.json` | Multi-step progress, System 2 only | ## Core rules - Produce at most one meaningful environment action per tick. - Actions must match available actions in `state/observation.txt`. - Never invent unavailable actions. - After `codegen` returns `success` or `in_progress`, stop with `done`. - Do not observe again in the same tick. - `done` means the current tick is complete, not necessarily that the intention is complete. ## Priority override If `safety`, `energy`, or `satiety` is below `0.2`, address the critical need first when possible. Read levels from `state/needs.json` when present; otherwise use `state/emotion.json` → `needs` or need hints in `state/observation.txt`. Priority: 1. Safety 2. Energy 3. Satiety If an existing multi-step plan can be resumed later, mark it as `interrupted`, not `failed`. ## Mode selection Default to System 1. Use System 1 when the next action is routine, obvious, or urgent. Use System 2 only when the goal requires multiple ticks, has uncertainty, involves conflict, or needs step tracking. ## System 1 For simple goals: 1. Read intention and observation. 2. Choose one available action. 3. Call `codegen` once. 4. Stop with `done`. Do not create `state/plan_state.json`. ## System 2 For multi-step goals: 1. Read or create `state/plan_state.json`. 2. Keep the plan within 6 steps. 3. Execute only the current step via `codegen`. 4. Update plan status if needed. 5. Stop after one action. Advance `current_step` only when the latest observation confirms the previous step is complete. ## Failure handling If `codegen` fails: - Retry at most once with a clearly different available action. - If still blocked, stop with `done` and record the failure. - After 3 consecutive failures on the same step, mark the plan as `failed`. Self-check (optional): `python scripts/validate_plan_state.py state/plan_state.json` For details, use: - `references/decision_rules.md` - `references/plan_state_schema.json` - `references/examples.md`
Running prompts needs a free account.
Sign in and we'll stream the response from Claude Sonnet 4.6 right here — no config needed for the platform models.
Execute the current intention via codegen. Use when acting on state/intention.json against the environment.
# Plan Convert the current intention into exactly one concrete environment action for this tick via `codegen`. This skill is a tick-level action controller. It does not complete the whole goal at once. After one meaningful action, stop. ## Inputs | File | Use | | ----------------------- | ----------------------------------------------- | | `state/intention.json` | Current goal or intention | | `state/observation.txt` | Available actions, context, affordances | | `state/needs.json` | Optional authoritative need levels | | `state/emotion.json` | Optional `needs` snapshot if needs.json missing | | `state/plan_state.json` | Multi-step progress, System 2 only | ## Core rules - Produce at most one meaningful environment action per tick. - Actions must match available actions in `state/observation.txt`. - Never invent unavailable actions. - After `codegen` returns `success` or `in_progress`, stop with `done`. - Do not observe again in the same tick. - `done` means the current tick is complete, not necessarily that the intention is complete. ## Priority override If `safety`, `energy`, or `satiety` is below `0.2`, address the critical need first when possible. Read levels from `state/needs.json` when present; otherwise use `state/emotion.json` → `needs` or need hints in `state/observation.txt`. Priority: 1. Safety 2. Energy 3. Satiety If an existing multi-step plan can be resumed later, mark it as `interrupted`, not `failed`. ## Mode selection Default to System 1. Use System 1 when the next action is routine, obvious, or urgent. Use System 2 only when the goal requires multiple ticks, has uncertainty, involves conflict, or needs step tracking. ## System 1 For simple goals: 1. Read intention and observation. 2. Choose one available action. 3. Call `codegen` once. 4. Stop with `done`. Do not create `state/plan_state.json`. ## System 2 For multi-step goals: 1. Read or create `state/plan_state.json`. 2. Keep the plan within 6 steps. 3. Execute only the current step via `codegen`. 4. Update plan status if needed. 5. Stop after one action. Advance `current_step` only when the latest observation confirms the previous step is complete. ## Failure handling If `codegen` fails: - Retry at most once with a clearly different available action. - If still blocked, stop with `done` and record the failure. - After 3 consecutive failures on the same step, mark the plan as `failed`. Self-check (optional): `python scripts/validate_plan_state.py state/plan_state.json` For details, use: - `references/decision_rules.md` - `references/plan_state_schema.json` - `references/examples.md`