v0.3.0-rc.2: manifesto / needs ladder L0-L10 #25

2026-05-27T17:50:43+03:00

halofourteen commented

2026-05-27 17:50:43 +03:00

(Migrated from github.com)

What and why

Pre-v0.3.0, the bot had no internal hierarchy of needs. The curriculum
produced a single "next milestone" but no priority structure. So when
the bot was wedged with no pickaxe, it kept dispatching explore.far
instead of recognising "I need wood → planks → pickaxe first". Pi could
see this in reflection logs but had no internal-state language to
express "L2 tools_wood is unmet".

This rc gives the bot an explicit 11-level Maslow-like needs ladder:

L0  alive          HP>5, food>0, not lava, not panic-near hostile
L1  food           ≥6 food items in inventory
L2  tools_wood     wooden_pickaxe + wooden_axe + wooden_sword
L3  shelter_basic  bed placed nearby or in inventory
L4  tools_stone    stone-tier triplet
L5  armor_basic    any chestplate (pursue=null for now)
L6  food_security  ≥16 food items
L7  tools_iron     iron-tier triplet
L8  armor_iron     iron chestplate (pursue=null for now)
L9  village_seed   bed + chest nearby
L10 village_full   global goal (always falls through)

Each reflex tick picks the lowest unsatisfied pursuable need and
dispatches a concrete skill toward it. Needs whose pursue() returns
null (armour, currently — no craft.leather-* skills yet) gracefully
skip without stalling the bot.

Changes

runtime/manifesto/needs.js (new) — 11-need catalogue with
detect(snapshot) and pursue(snapshot) per need.
runtime/manifesto/state.js (new) — pickActiveNeed(snapshot) walks
the ladder, returns {need, skillId, args, blockedNeeds}. 3s cache.
Validates skillId against the live registry (rc.1 plumbing) — the
manifesto literally cannot ship a hallucinated skill name.
runtime/reflex.js — curriculumReflex consults manifesto first.
Curriculum plan becomes the fallback. Tests can pass
ctx.disableManifesto = true to exercise the curriculum branch
in isolation (existing tests do that; new tests exercise manifesto-on).
runtime/coach/reflect.js — Pi self-reflection prompt now includes
"activeNeed: L2 tools_wood → gather.logs (Деревянные орудия)" so Pi
advises at the right level.

Behavioural diff (illustrative)

Situation	v0.2.x dispatch	v0.3.0-rc.2 dispatch
food=20, no inventory, curriculum=logs	`gather.logs`	`survive.acquire-food` (L1)
Full wood tools, no bed, no stone	`gather.logs` (loop)	`gather.wool` (L3)
Full wood + stone tools, no chestplate	curriculum plan	armor skipped → L6 food
HP=4, creeper@3m, has tools, has bread	curriculum plan	`survive.flee` (L0)

Test plan

npm test — 315 green (was 279 on rc.1, +36 new):
- 24 unit tests in manifesto/needs.test.js (per-need)
- 10 unit tests in manifesto/state.test.js (ladder walk, L0 takeover,
  skip-on-null, caching)
- 2 integration tests in reflex.test.js
Post-merge: deploy, watch
```
sqlite3 state/.../knowledge.db \
  "SELECT applied_count, source, text FROM lessons
   WHERE applied_count > 0 ORDER BY applied_count DESC LIMIT 20;"
```
Expect Pi-coach/Pi-reflect applied_count to start growing as the
bot makes ladder progress (each progression generates valid
prefer_skill suggestions Pi can build on).

Followup (rc.3)

Event-driven awareness layer: bot.on('move') jump-detect,
bot.on('health') plunge-detect, bot.on('entitySpawn') hostile-add
AbortSignal plumbing through long skills (gather.logs, explore.far,
recovery.tunnel-out, survive.pillar-up) so they preempt within
~100ms instead of waiting for a tick boundary
Wire fast-advisor (rc.1 scaffold) into the awareness layer

See dev/v0.3.0/PLAN.md for the full v0.3.0 design.

🤖 Generated with Claude Code

## What and why Pre-v0.3.0, the bot had no internal hierarchy of needs. The curriculum produced a single "next milestone" but no priority structure. So when the bot was wedged with no pickaxe, it kept dispatching `explore.far` instead of recognising "I need wood → planks → pickaxe first". Pi could *see* this in reflection logs but had no internal-state language to express "L2 tools_wood is unmet". This rc gives the bot an explicit 11-level Maslow-like needs ladder: ``` L0 alive HP>5, food>0, not lava, not panic-near hostile L1 food ≥6 food items in inventory L2 tools_wood wooden_pickaxe + wooden_axe + wooden_sword L3 shelter_basic bed placed nearby or in inventory L4 tools_stone stone-tier triplet L5 armor_basic any chestplate (pursue=null for now) L6 food_security ≥16 food items L7 tools_iron iron-tier triplet L8 armor_iron iron chestplate (pursue=null for now) L9 village_seed bed + chest nearby L10 village_full global goal (always falls through) ``` Each reflex tick picks the **lowest unsatisfied pursuable** need and dispatches a concrete skill toward it. Needs whose `pursue()` returns `null` (armour, currently — no `craft.leather-*` skills yet) gracefully skip without stalling the bot. ## Changes - `runtime/manifesto/needs.js` (new) — 11-need catalogue with `detect(snapshot)` and `pursue(snapshot)` per need. - `runtime/manifesto/state.js` (new) — `pickActiveNeed(snapshot)` walks the ladder, returns `{need, skillId, args, blockedNeeds}`. 3s cache. **Validates skillId against the live registry** (rc.1 plumbing) — the manifesto literally cannot ship a hallucinated skill name. - `runtime/reflex.js` — `curriculumReflex` consults manifesto first. Curriculum plan becomes the fallback. Tests can pass `ctx.disableManifesto = true` to exercise the curriculum branch in isolation (existing tests do that; new tests exercise manifesto-on). - `runtime/coach/reflect.js` — Pi self-reflection prompt now includes `"activeNeed: L2 tools_wood → gather.logs (Деревянные орудия)"` so Pi advises at the right level. ## Behavioural diff (illustrative) | Situation | v0.2.x dispatch | v0.3.0-rc.2 dispatch | |---------------------------------------------|------------------------|-----------------------------| | food=20, no inventory, curriculum=logs | `gather.logs` | `survive.acquire-food` (L1) | | Full wood tools, no bed, no stone | `gather.logs` (loop) | `gather.wool` (L3) | | Full wood + stone tools, no chestplate | curriculum plan | armor skipped → L6 food | | HP=4, creeper@3m, has tools, has bread | curriculum plan | `survive.flee` (L0) | ## Test plan - [x] `npm test` — 315 green (was 279 on rc.1, +36 new): - 24 unit tests in `manifesto/needs.test.js` (per-need) - 10 unit tests in `manifesto/state.test.js` (ladder walk, L0 takeover, skip-on-null, caching) - 2 integration tests in `reflex.test.js` - [ ] Post-merge: deploy, watch ``` sqlite3 state/.../knowledge.db \ "SELECT applied_count, source, text FROM lessons WHERE applied_count > 0 ORDER BY applied_count DESC LIMIT 20;" ``` Expect Pi-coach/Pi-reflect `applied_count` to start growing as the bot makes ladder progress (each progression generates valid prefer_skill suggestions Pi can build on). ## Followup (rc.3) - Event-driven awareness layer: `bot.on('move')` jump-detect, `bot.on('health')` plunge-detect, `bot.on('entitySpawn')` hostile-add - AbortSignal plumbing through long skills (gather.logs, explore.far, recovery.tunnel-out, survive.pillar-up) so they preempt within ~100ms instead of waiting for a tick boundary - Wire fast-advisor (rc.1 scaffold) into the awareness layer See [`dev/v0.3.0/PLAN.md`](dev/v0.3.0/PLAN.md) for the full v0.3.0 design. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

halofourteen commented

2026-05-27 18:09:39 +03:00

(Migrated from github.com)

Superseded by single v0.3.0 PR

Pull request closed

This pull request cannot be reopened because the branch was deleted.

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: xmatic/pepa-pi-bot#25