npm - @oneie/claude - Versions diffs - 0.5.0 → 0.7.0 - Mend

@oneie/claude 0.5.0 → 0.7.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (2118) hide show

package/rules/design.md CHANGED Viewed

@@ -19,8 +19,8 @@ default palette so wrong colors emit no CSS. Don't fight the enforcement.
 | Token | Use for |
 | --- | --- |
-| `background` | Card surfaces, sidebars, page-level panels |
-| `foreground` | Inner content rectangles inside cards |
+| `background` | Page canvas — the primary working surface, behind cards and sidebar |
+| `foreground` | Cards, sidebar, elevated panels — surfaces that float above the canvas |
 | `font` | All body text |
 | `primary` | Main CTAs, brand accents, focus rings |
 | `secondary` | Supporting actions, secondary buttons |
@@ -40,11 +40,17 @@ Polish constants (baked, not exposed): `--radius-sm` 6px · `--radius-md` 10px
 | Level | Surface | Where |
 | --- | --- | --- |
-| L0 page | `--color-page` | `<body>`, full-bleed shell |
-| L1 card | `--color-background` | Cards, sidebar, popovers, dropdowns |
-| L2 content | `--color-foreground` | Card body, inputs, code blocks |
+| L0 page | `--color-page` | `<body>`, full-bleed shell — derived: background + 4% font |
+| L1 canvas | `--color-background` | Main content area — the page canvas |
+| L2 chrome | `--color-foreground` | Sidebar, cards, panels — float above the canvas |
-Sidebar = L1. Inputs = L2. There is no L3. A card header/footer shares the card surface; never tint them separately.
+In dark mode: background (10%) is deeper than foreground (13%) — chrome panels rise above the canvas. In light mode: background (93% gray) is muted, foreground (100% white) is bright — panels shine above the canvas.
+Inputs sink back to `background` (the canvas level) when inside a `foreground` panel — they feel recessed into the panel surface.
+There is no L3. A card header/footer shares the card surface; never tint them separately.
+> **Pending code rename:** the component code currently applies `bg-background` to cards and `bg-foreground` to card interiors — inverse of this model. A mechanical rename (`bg-background` ↔ `bg-foreground` on card/sidebar elements) is tracked separately. When doing that cycle, grep `bg-background` on article/aside/nav elements and `bg-foreground` on inner content divs.
 ---
@@ -87,6 +93,7 @@ The `--color-{border,muted,ring}` CSS vars exist (defined in `Layout.astro`) but
 - ❌ Mixing icon sets. Lucide only — via `<Icon>` / `<IconBadge>` (in `web/src/components/ui/`).
 - ❌ Inline SVG icons in React components. Import from `lucide-react` and wrap.
 - ❌ Unicode icon glyphs (☀ ☾ ▾ ✓ ✗). They render differently across OSes — use lucide.
+- ❌ `text-foreground` as a text color. `foreground` is the L2 panel *surface* token, not a contrast label — using it for text renders text the same color as its own background (invisible, and it flips wrong between light/dark). Use `text-font` for body text, `text-on-{primary|secondary|tertiary}` for text on a brand fill. **Exception:** the `bg-font text-foreground` Inverse button — surface-colored text on a font-colored fill is correct. Fine as-is (not policed here): `text-muted-foreground`/`text-card-foreground`/`text-popover-foreground`/`text-accent-foreground` (Layout.astro aliases them to `font`-derived colors) and shadcn's `text-{primary,secondary,destructive,sidebar}-foreground` contrast labels on brand fills (Button/Badge/Checkbox — they resolve via globals.css).
 ---
@@ -149,7 +156,7 @@ In Astro pages where you can't easily import React, use inline SVG with `viewBox
 ### Form fields
-Inputs use `background` (sunken), not `foreground` (raised). The card body is `foreground`; inputs sink back to `background` so they stand out as interactive surfaces against the body.
+Inputs use `background` (canvas level) inside a `foreground` panel — they appear recessed/sunken against the card surface. This is the correct spatial read: you're looking through the panel (foreground) into the canvas (background).
 ```tsx
 <div className="flex flex-col gap-1.5">
@@ -196,11 +203,21 @@ Three layers, all automatic:
 2. **PostToolUse hook** — `.claude/hooks/design-check.sh` greps every Write/Edit
    to `one.ie/web/**/*.{tsx,astro,css}` for banned patterns. Exit 2 on violation feeds
    the diff back to Claude as a tool error; the model self-corrects next turn.
-3. **This rule** — auto-loaded on the same files via `.claude/settings.json`,
-   so the constraints are in context before the first character is written.
+3. **This rule** — auto-loaded on the same files via the `paths:` frontmatter at
+   the top of this file (not `settings.json`, which wires hooks only), so the
+   constraints are in context before the first character is written.
 The hook allowlists `Layout.astro` (token source) and `design.astro` (showcase).
-There is no opt-out for other files.
+**Layers 2 and 3 are edit-triggered, so pre-existing violations survive.** The
+hook is PostToolUse: it only sees a file when that file is written. Nothing
+sweeps the tree, and layer 1 structurally cannot catch a hex string inside a JS
+object — it nulls Tailwind palette *classes*, not literals. So a file untouched
+since this rule landed can carry banned patterns indefinitely.
+`src/components/paths/PathGraph.tsx` did: it carried hex literals at lines 27,
+28, 29 and 55 until they were swept to `var(--color-{success,secondary,destructive})`.
+Treat "the hook is green" as "nobody has edited the offender", not as "the tree
+is clean".
 ---

package/rules/documentation.md CHANGED Viewed

@@ -2,7 +2,32 @@
 Apply to all TODOs and `/do` workflows.
-**Principle:** Document BEFORE implementation (W2), edit alongside code (W3), verify consistency (W4).
+**Principle:** Docs are the spec, written **first**. The feature's doc set is authored at the **DOCS** stop — *before* any code — in the reader's language, and the tests are drawn from it. The docs are then the **acceptance oracle**: PROVE validates the shipped thing against them, and TEACH reconciles any drift so every doc ends true.
+**Upstream of the doc set sits the promise** — `text/<slug>.md`, the genesis contract, written like the statement of work an engineering agency signs with a client, with **four** machine-read manifests: `deliverables:` (the exhaustive schedule of work — every item the client receives, each with its own `accept:` check; not listed = not promised, listed = ships or the promise settles broken, acceptance indivisible; `assumes:` lists client-side dependencies, green at the making), `proof:` (the acceptance test — derived, not authored: the `&&`-join of every `accept:`; it becomes the todo's `outcome:`, the TEST assertion, the PROVE oracle, and the on-chain settlement condition; the schedule ⇄ proof law is gated by `.claude/scripts/do-promise-lint.sh`), `derives:` (which artifacts the promise spawns — `/do` reads it to know which doc stops to backfill), and `world:` (the runtime a kept promise puts into the substrate — lifecycle moves, workflow steps, agents + `subscribes:` tags, skills, task tags, tracking marks/warns, routing paths, views). Every `world:` entry is **presence-checked like an artifact**: missing agent → `template-agent.md` cycle + `subscriptions:register` · missing skill → `/skill-creator` · missing workflow/lifecycle stage → a todo cycle · missing view kind/route (`one.ie/web/src/lib/views/registry.ts` + the order-free traversal) → a todo cycle · present → skip. The covenant behind the mechanism: `text/promises.md`. Every doc reconciles upward to the promise — a doc claim that contradicts it is a bug in the doc. Full contract: `.claude/commands/do.md` § "PROMISE first".
+Within a build cycle this still maps to three phases: **W2** declares which docs this cycle touches (the spec already exists; W2 lists the deltas), **W3** edits docs alongside code in the same wave, **W4** verifies consistency *and truth* against the shipped behaviour. The difference from before: the full doc set is not seeded-then-refined — it is written for real up front, and the cycle keeps it in sync and proves it true.
+## Promises grow with delivery (always update the promise with what shipped)
+**When you ship a feature that extends an already-kept promise — a new view, capability, or surface under the same banner — you MUST append it to that promise as a new `deliverable:` with its own `accept:` check, and extend the `proof:` `&&`-join to include that check.** This is not optional polish; it is what keeps the promise a record the system can depend on. A promise that silently omits shipped work is a promise that lies about the delivery, and the router that sells the next signal to the proven path is reading a stale contract.
+The covenant's freeze (clause 4, `text/promises.md`) permits this because it is a **monotonic ratchet** — the same "can't regress" DNA `/do` runs on:
+- **Allowed (do it every time):** append a deliverable whose `accept:` is **already green**, and add that exact `accept:` string to `proof:`. The join stays exit-0; the record only rises. No new `terms_hash` needed — nothing that was promised changed.
+- **Forbidden (needs a *new* promise):** anything that would make the existing proof go **red** — weakening a check, removing a line, reversing a decision, or a broken `assumes:`. That's a regression, not growth.
+Checklist when extending a kept promise `text/<slug>.md`:
+1. Add the `- item:` + `accept:` row under `deliverables:` (group fulfilled follow-ons under a dated comment).
+2. Append the identical `accept:` body to `proof:` (the schedule ⊆ proof law — `do-promise-lint.sh` enforces verbatim inclusion).
+3. Add any new `world:` entries (a new view → `views:`, a new workflow step → `workflow:`).
+4. Add the human-readable row to the deliverables table in the prose, and reconcile any now-stale status claim (e.g. an "X is still red / not shipped" line for the thing you just shipped).
+5. Run `bash .claude/scripts/do-promise-lint.sh <slug>` — it must report `schedule ⊆ proof, contract well-formed` **and** every `assumes:` green. Fix any stale `assumes:` path while you're there (a moved file makes the whole contract un-signable).
+The same discipline applies to the propagate matrix below: the promise is the first row every close reconciles, before its derived docs.
+**Format:** any ASCII diagram written into a doc follows `text/CLAUDE.md § ASCII diagrams` — max ~60 chars wide, vertical flow over multi-column tables. A diagram that's fine in a wide terminal wraps into garbage in the narrow panels these docs actually get read in.
 ---
@@ -19,9 +44,9 @@ Before any code is written, explicitly list which docs will change:
 - `text/feature.md` — {purpose}
 **Docs modified:**
-- `text/dictionary-plan.md` — add term {name}
+- `text/dictionary.md` — add term {name}
 - `text/routing-plan.md` — update loop L{N}
-- `text/rubrics-plan.md` — add dimension {name}
+- `text/rubrics.md` — add dimension {name}
 **Schema changes:**
 - New TypeDB entities, D1 migrations, TypeQL functions
@@ -33,18 +58,22 @@ Before any code is written, explicitly list which docs will change:
 ### **W3 — Edit Phase: Update Docs + Code in Parallel**
-**For every code file edited, edit the corresponding doc:**
+**Docs are a required W3 layer** — W2 declares which docs will change in `.w2-doc-plan.json`, and W3 applies those changes in the same W3a wave as code edits. Docs edits are spawned as parallel agents alongside code, not sequentially after.
+**For every code file edited, edit the corresponding doc (always in W3a, never W3b):**
+| Code File | Related Doc | What to add/update |
+|-----------|-------------|---|
+| `src/types/foo.ts` (new file) | `text/dictionary.md` | Add term: `Foo:` definition + field breakdown |
+| `src/pages/api/foo.ts` (new route) | `text/lifecycle.md` | Add stage: when agent/operator uses this route, what they get |
+| `src/components/foo/Foo.tsx` (new component) | Feature doc (`text/<feature>-plan.md`) | Add section: component purpose, props, slots, states |
+| `migrations/0042_foo.sql` (schema change) | `text/one-ontology.md` | Add/update entity: new type or attribute; which actor reads it |
+| `packages/sdk/src/index.ts` (new export) | `text/dictionary.md` | Ensure exported types are documented |
+| Feature-wide changes | `text/<feature>-plan.md` | Update: add lifecycle section, API shapes, example usage, when/why to use |
-| Code File | Related Doc |
-|-----------|-------------|
-| `nanoclaw/src/types.ts` | `text/dictionary-plan.md` — add new type terms |
-| `src/engine/world.ts` | `text/DSL-plan.md` — update signal grammar |
-| `src/engine/loop.ts` | `text/routing-plan.md` — update L1-L7 loops |
-| `src/pages/api/*.ts` | `text/lifecycle-plan.md` — update lifecycle stage |
-| `src/schema/*.tql` | `text/one-ontology.md` — update dimension/entity |
-| Feature-specific file | `text/feature.md` — update feature spec |
+**Pattern:** If you're adding a new field to a TypeScript interface, add it to the doc that defines that interface. If you're adding a new route, document when an operator/agent calls it and what they get back. If you're changing data structure, update the type definition in the dictionary.
-**Pattern:** If you're adding a new field to a TypeScript interface, add it to the doc that defines that interface. If you're changing a loop, update the loop diagram in the doc.
+**Critical rule:** Never commit code without corresponding doc edits in the same W3a wave. The W3a message spawns code-edit agents AND doc-edit agents together. A code change with no doc update is an incomplete feature and fails W4.
 ---
@@ -96,79 +125,12 @@ These six docs are the **source of truth** and must stay in sync with code:
 | **one-ontology.md** | 6 dimensions, actor/group/thing/path/signal/hypothesis | Type system changes |
 | **routing.md** | L1-L7 loops, signal flow, priority formula | Loop/routing changes |
 | **lifecycle.md** | Agent journey stages, revenue flow | Lifecycle/economic changes |
-| **rubrics.md** | Code rubric (security/stability/simplicity/speed) + agent rubric (fit/form/truth/taste) | Rubric changes or new dimensions |
+| **rubrics.md** | Code rubric (the design target — eight dims) + agent rubric (fit/form/truth/taste). Live weights are data at `.claude/scripts/rubric-weights.json`; the gap between the two is `text/rubric-migration-plan.md` | Rubric changes or new dimensions |
 **Feature-specific docs** (rich-messages.md, webhooks.md, etc.) are updated per feature. Always link back to the six core docs.
 ---
-## The Workflow Integration
-### Hook: Pre-W2 — Load Source Docs
-When a TODO is created, auto-load relevant source docs:
-```
-/do {name}-todo.md
-├─ [Hook] Load: dictionary.md, DSL.md, rubrics.md
-├─ [W1] Recon code (reads files)
-├─ [W2] Decide code + **document changes needed in docs**
-│       (explicit list: what terms change, what loops change, what's new)
-├─ [W3] Edit code (spawns Sonnet agents)
-│       + [Hook] Suggest: "Also edit X doc for {Y change}"
-└─ [W4] Verify code + **verify docs match** (cross-check task)
-```
-### Hook: Post-W4 — Validate Doc Consistency
-After W4 completes, auto-run consistency checks:
-```bash
-# Terminology consistency
-grep -r "old-term" text/ | flag if > 0 (not on deadname list)
-# Cross-references
-markdown-link-check text/*.md
-# Metaphor alignment (if touching signal/path semantics)
-grep -l "signal\|strength\|resistance" src/changed-files.txt | xargs -I {} \
-  check if metaphors.md still applies
-# Rubric scoring (if W4 added dimensions)
-grep "rubric\|dimension" src/changed-files.txt | flag for rubrics.md review
-```
----
-## Example: rich-messages-todo (Retrospective)
-**What should have happened (W2):**
-```markdown
-### Documentation Updates (W2)
-**New docs:**
-- None (extends existing types)
-**Docs modified:**
-- `text/dictionary-plan.md` — add "Rich Messages" section, explain `data.rich` convention
-- `text/rich-messages.md` — add PaymentMetadata type, Payment flow section
-- `text/template-todo.md` — add "Documentation Updates (W2 + W4)" section
-**Schema changes:**
-- `nanoclaw/src/types.ts` — RichMessage.thread, RichMessage.payment added (Cycle 1 & 3)
-- `migrations/0010_payment_columns.sql` — new D1 columns for payment tracking
-```
-**What actually happened:**
-- ✅ Docs updated (rich-messages.md, dictionary.md, template-todo.md) — **AFTER W4**
-- ⚠️ Should have been planned in W2, preventing surprises in W3
-**Lesson:** Next TODO follows: document in W2, edit in W3, verify in W4. No surprises.
----
 ## Rules for `/do` (The Command)
 When running `/do {name}-todo.md`, the workflow enforces:
@@ -176,7 +138,7 @@ When running `/do {name}-todo.md`, the workflow enforces:
 1. **W2 must explicitly call out doc changes** — no surprise doc edits in W3
 2. **W3 spawns edit agents for both code AND docs** — parallel edits
 3. **W4 includes doc consistency check** — not just code tests
-4. **The code rubric (security/stability/simplicity/speed) applies to docs too** — simplicity means fewer words; stability means no broken links; security means no sensitive data exposed; speed means lean docs = fewer tokens to read
+4. **The code rubric applies to docs too**, reading each axis for prose — simplicity means fewer words; stability means no broken links and accurate references; security means no sensitive data exposed; integration means every doc is linked from its family and lies about nothing shipped; speed means lean docs = fewer tokens to read. (Axis names and weights: `.claude/scripts/rubric-weights.json`. Note the doc reading of *stability* — links and references — is not the code reading — error handling and closed loops; same name, two criteria, which is exactly why the eight-dim migration splits it.)
 ---
@@ -190,6 +152,8 @@ When running `/do {name}-todo.md`, the workflow enforces:
 | Trigger (detected from W3 diff) | Target doc | Edit |
 |---|---|---|
+| feature extends an already-kept promise | that `text/<slug>.md` promise | **first row, always** — append the fulfilled deliverable + its green `accept:`, extend `proof:`, add `world:` entries, then `do-promise-lint.sh <slug>` must pass (see § Promises grow with delivery) |
+| slug has `text/<slug>-agents.md` (a verification walk exists) | `.claude/scripts/do-walk.sh <slug> --agents` | run it — must exit 0; a promise extension that added a deliverable but no matching walk stop is an incomplete close, append the stop to both `-humans.md` and `-agents.md` before the cycle closes |
 | any change | source-of-truth doc(s) from plan frontmatter | always — even if only touch-verified |
 | 6-dim, L1-L8, or locked-rule change | root `CLAUDE.md` | edit the relevant section |
 | directory contract change (new component family, new pattern) | nearest `CLAUDE.md` (e.g., `one.ie/web/src/components/CLAUDE.md`) | append / edit section |
@@ -202,17 +166,27 @@ Anchor mismatch → log to `text/improvements.md` for manual W2 next cycle (no h
 **Learnings entry** (append to `text/learnings.md` on each close):
 ```
-- YYYY-MM-DD · cycle N · wave N|gate · {one sentence} · rubric=0.NN · source=w1|w2|w3|w4|cycle
+- YYYY-MM-DD · cycle N · wave N|gate · {one sentence} · rubric=0.NN [sec.NN simp.NN struct.NN speed.NN integ.NN reuse.NN test.NN ux.NN] · source=w1|w2|w3|w4|cycle
+The `[...]` eight-field vector is self-labeling and appended, never replacing the
+composite — see `text/rubrics.md` § Code Rubric for the eight dimensions and the
+per-field scoring rules. A close with no per-dimension breakdown to hand may omit
+the vector (old form still parses; counts toward the composite trend only).
 ```
 W4 code rubric applies to docs too: security (no sensitive data), stability (no broken links, accurate references), simplicity (fewer words, no bloat), speed (lean = fewer tokens to read).
 ### W4 doc-sync hard gate (zero LLM)
-W4 must pass three bash checks before the rubric runs. Detail in `.claude/commands/do.md` W4 section. Summary:
+W4 must pass four bash checks before the rubric runs. Detail in `.claude/commands/do.md` W4 section. Summary:
 1. **stale-name check** — every old identifier in W3's rename list has 0 hits in `**/*.md`
-2. **broken-link check** — `markdown-link-check` on every touched doc returns 0 broken
+2. **broken-link check** — `markdown-link-check` on every touched doc returns 0 broken. **It is not installed** (`command -v markdown-link-check` → nothing), so this check cannot pass today. Report it `n/a`, never `ok` — an absent tool is an unrun check, not a green one. Install it or replace it with a resolve-every-relative-link grep; do not let it silently vanish.
 3. **contract-staleness check** — every dir whose code was touched has `CLAUDE.md` mtime ≥ newest code mtime
+4. **shipped-status check** — the just-built slug's own `text/<slug>-plan.md` / `-docs.md` must not still claim the feature is unbuilt:
+   ```bash
+   grep -liE 'design doc \(not an armed|not yet built|^not shipped|is the A2A follow-on|durable.*follow-on, not' text/<slug>-plan.md text/<slug>-docs.md 2>/dev/null
+   ```
+   A hit on the plan that was *just built* is a doc that lies about what shipped → reconcile the phrase (or, if it genuinely refers to a separate future item, reword so it can't be confused with the shipped thing). This is the recurring drift (3/3 runs eve-C5 · remote-durable · remote-suspend): a feature-extension cycle ships the capability but leaves "X is the follow-on / not shipped" in its own plan/docs/code-headers. Catches the status-claim drift that #1 (names) and #2 (links) don't.
 Any fail → cycle does NOT close. No rubric until docs match code.

package/rules/engine.md CHANGED Viewed

@@ -68,8 +68,8 @@ dimension scores. These are the deterministic signals that calibrate path streng
   buildMs: 22995,
   deployMs: { web: 13705, sync: 8249, channels: 9163, api: 17391 },
   health: { web: 292, sync: 270, channels: 270 },
-  // /do code rubric — security/stability/simplicity/speed, gate ≥ 0.65
-  rubric: { security: 0.92, stability: 0.85, simplicity: 0.90, speed: 0.80 }
+  // /do code rubric — security/stability/simplicity/integration/speed, gate ≥ 0.65
+  rubric: { security: 0.92, stability: 0.85, simplicity: 0.90, integration: 0.88, speed: 0.80 }
 }
 ```
@@ -80,7 +80,7 @@ sandwich applied to every automation skill.
 **Where it shows up:**
 - `/do` — W4 reports tests passed/total, reconcile canons, rubric scores
-- `/close` — reports rubric score (security + stability + simplicity + speed, composite ≥ 0.65) + learnings entry
+- `/close` — reports rubric score (security + stability + simplicity + integration + speed, composite ≥ 0.65) + learnings entry + promise settle (kept/broken/dissolved/none)
 - `/sync` — reports tasks synced, hash-delta, KV writes
 - `/deploy` — reports W0 results, build ms, per-service deploy ms, health
@@ -120,7 +120,20 @@ world.restore(snapshot)     // rehydrate
 ```
 Over HTTP the same verbs are `SubstrateClient` methods (`packages/sdk/src/client.ts`).
-`harden` is exposed via `signal("learning:know")`.
+`harden` is the one verb with **no general wire**. `learning:know` is in no
+registry (`grep -c "learning:" packages/sdk/src/receivers.ts` → 0) and has no
+resolver; `client.know()` is `@deprecated` and returns `{}`. The only shipped
+harden receiver is **`chat:harden`** (`one.ie/web/src/lib/resolvers/entity.ts:566`),
+which crystallises a resolved chat thread into a dim-6 learning and requires an
+`entityId` — so it does not cover free-standing claims. Do not write
+`signal("learning:know")` into new code or docs; it fails silently.
+**Promise paths.** `promise:<slug>→proof` is a first-class weighted path
+family. At close, `.claude/scripts/do-promise-settle.sh` re-runs the
+promise's `proof:` and `mark`s the path with the W4 composite (kept) or
+`warn`s it (broken). Build-time and run-time reputation are one ledger —
+cycle composites land on the same weighted paths run-time routing reads.
 ---
@@ -146,6 +159,61 @@ and warns the edge (Rule 1). Dissolve is an outcome, not an exception.
 ---
+## Rules 4–6 (from The Shape of the System)
+### Rule 4 — Deadlines + Ceilings
+**Every external call has a deadline. Every caller-controlled loop has a ceiling.**
+```typescript
+// GOOD — AbortController on every fetch to an uncontrolled boundary
+const controller = new AbortController()
+const id = setTimeout(() => controller.abort(), TIMEOUT_MS)
+try {
+  const res = await fetch(url, { signal: controller.signal })
+} finally {
+  clearTimeout(id)
+}
+// BAD — no deadline = the caller's failure becomes your hang
+const res = await fetch(url)
+```
+Ungated retries and uncapped allocations (sockets, threads, recursion) are DoS bugs — fixed on sight, no profiling required. Applies to: TypeDB calls, KV reads, chain RPCs, price oracles, any `fetch` to an external host.
+### Rule 5 — Drain Before Exit
+**CF Workers drain before returning.** Stop accepting → drain in-flight within deadline → flush durable writes → return.
+```typescript
+// pattern for Durable Objects and long-running handlers
+ctx.waitUntil(drainAndFlush())
+return new Response('accepted', { status: 202 })
+```
+Half-finished work silently corrupts downstream. The OS reclaims memory; it never reclaims meaning. Signal loss between "credits granted" and "signal emitted" is this failure mode.
+### Rule 6 — Monotonic Durations
+**Durations use a monotonic source. Wall clock is for display and TTL anchoring at write time only.**
+```typescript
+// GOOD — duration from performance.now() (monotonic)
+const start = performance.now()
+// ... work ...
+const elapsedMs = performance.now() - start
+// GOOD — TTL anchored once at write time (wall clock OK here)
+await KV.put(key, value, { expirationTtl: TTL_SECONDS })
+// BAD — wall clock for ordering or mid-flight deadline check
+if (Date.now() > deadline) throw new Error('expired') // skews under NTP
+```
+Wall clock can jump or skew under NTP corrections. Never use it to compare event ordering across Workers or to check a deadline computed from a prior `Date.now()` call.
+---
 ## Verb Contracts
 Each of the 6 verbs has a **pre / post / invariant** contract. Spec lives in `text/contracts.md` — loaded only when editing a verb implementation, not every session.

package/rules/react.md CHANGED Viewed

@@ -114,16 +114,16 @@ export default function AgentCard() { ... }
 ## With Substrate
 ```tsx
-import { colony } from '@/engine/substrate'
+import { createWorld } from '@/lib/pheromone'
 export function SwarmView() {
-  const [net] = useState(() => colony())
+  const [net] = useState(() => createWorld())
   const [highways, setHighways] = useState([])
   useEffect(() => {
     const interval = setInterval(() => {
       net.fade(0.1)
-      setHighways(net.highways(10))
+      setHighways(net.highways(undefined, 10))
     }, 1000)
     return () => clearInterval(interval)
   }, [net])

package/rules/ui.md CHANGED Viewed

@@ -35,8 +35,12 @@ Receiver naming: `ui:<surface>:<action>` where surface is the component or page
 | `settings` | close, save, reset | `ui:settings:close`, `ui:settings:save` |
 | `memory` | reveal, forget, expand | `ui:memory:reveal`, `ui:memory:forget` |
 | `demo` | run, reset, next | `ui:demo:run`, `ui:demo:reset` |
+| `in:space` / `in:view` | the inbox axes — select, open | `ui:in:space:select`, `ui:in:view:open` |
+| `switcher` | group switching | `ui:switcher:open`, `ui:switcher:select` |
+| `members` | add, remove, delegate | `ui:members:add`, `ui:members:delegate` |
-Format: `ui:<surface>:<action>` — all lowercase, colon-delimited.
+Format: `ui:<surface>:<action>` — all lowercase, colon-delimited. A surface may
+nest (`in:space`), but the action is always the last segment.
 ---

package/scripts/__pycache__/do-rank.cpython-313.pyc ADDED Viewed

Binary file

package/scripts/browser-check.mjs ADDED Viewed

@@ -0,0 +1,63 @@
+#!/usr/bin/env node
+/**
+ * browser-check.mjs — DEPRECATED SHIM. Kept so existing callers keep working.
+ *
+ * The engine is now `.claude/scripts/chrome.mjs` (chrome-headless-shell via
+ * Playwright). This file maps the old flag set onto it and renames the report
+ * fields back to the shape `do-prove.sh` and `/browser` already parse.
+ *
+ * New work calls chrome.mjs directly — it has steps, network, eval, selectors,
+ * cookies and headers this shim does not expose.
+ *
+ * Old usage (unchanged):
+ *   node .claude/scripts/browser-check.mjs [url] [--send "msg"] [--screenshot] [--network]
+ *
+ * Old prerequisite (`npm install playwright` into /tmp) is GONE. chrome.mjs
+ * finds Playwright wherever this tree installed it, including from a linked
+ * worktree. The /tmp copy had already evaporated, which silently degraded every
+ * browser PROVE to a curl fallback.
+ */
+import { spawnSync } from 'node:child_process'
+import { dirname, join } from 'node:path'
+import { fileURLToPath } from 'node:url'
+const HERE = dirname(fileURLToPath(import.meta.url))
+const args = process.argv.slice(2)
+const url = args.find((a) => a.startsWith('http')) ?? 'http://localhost:4321'
+const sendMsg = args.find((_, i) => args[i - 1] === '--send') ?? null
+const doScreenshot = args.includes('--screenshot')
+const doNetwork = args.includes('--network')
+const argv = [join(HERE, 'chrome.mjs'), url]
+if (sendMsg) argv.push('--send', sendMsg)
+if (doScreenshot) argv.push('--screenshot', '/tmp/browser-check.png')
+if (doNetwork) argv.push('--network')
+const run = spawnSync(process.execPath, argv, { encoding: 'utf8', maxBuffer: 32 * 1024 * 1024 })
+if (run.status !== 0) {
+  process.stderr.write(run.stderr || 'chrome.mjs failed\n')
+  process.exit(run.status ?? 1)
+}
+let r
+try {
+  r = JSON.parse(run.stdout)
+} catch {
+  process.stdout.write(run.stdout)
+  process.exit(0)
+}
+console.log(JSON.stringify({
+  url: r.url,
+  httpStatus: r.httpStatus,
+  title: r.title ?? '(no title)',
+  jsErrors: r.jsErrors,
+  consoleErrors: r.consoleErrors,
+  chatRailBefore: r.chatRailBefore ?? null,
+  chatRailAfter: sendMsg ? (r.chatRailAfter ?? null) : undefined,
+  chatCapture: sendMsg ? (r.chatCapture ?? []) : undefined,
+  apiRequests: doNetwork ? (r.network ?? []).map((n) => ({ method: n.method, url: n.url })) : undefined,
+  screenshot: doScreenshot ? '/tmp/browser-check.png' : undefined,
+}, null, 2))

package/scripts/bun.lock ADDED Viewed

@@ -0,0 +1,27 @@
+{
+  "lockfileVersion": 1,
+  "configVersion": 1,
+  "workspaces": {
+    "": {
+      "name": "@oneie/claude-scripts",
+      "dependencies": {
+        "@anthropic-ai/sdk": "^0.100.1",
+      },
+    },
+  },
+  "packages": {
+    "@anthropic-ai/sdk": ["@anthropic-ai/sdk@0.100.1", "", { "dependencies": { "json-schema-to-ts": "^3.1.1", "standardwebhooks": "^1.0.0" }, "peerDependencies": { "zod": "^3.25.0 || ^4.0.0" }, "optionalPeers": ["zod"], "bin": { "anthropic-ai-sdk": "bin/cli" } }, "sha512-RANcEe7LpiLczkKGOwoXOTuFdPhuubS0i4xaAKOMpcqc55YO0mukgxppV7eygx3DXNjxWT6RYOLPyOy0aIAmwg=="],
+    "@babel/runtime": ["@babel/runtime@7.29.7", "", {}, "sha512-Nq8OhGWiZIZGV6hLHoyAKLLcJihP/xFeBMGJoUrxTX2psI8dCifzLhZISFb+VWS3wFMRDmCGw5R+dOySCqPLhw=="],
+    "@stablelib/base64": ["@stablelib/base64@1.0.1", "", {}, "sha512-1bnPQqSxSuc3Ii6MhBysoWCg58j97aUjuCSZrGSmDxNqtytIi0k8utUenAwTZN4V5mXXYGsVUI9zeBqy+jBOSQ=="],
+    "fast-sha256": ["fast-sha256@1.3.0", "", {}, "sha512-n11RGP/lrWEFI/bWdygLxhI+pVeo1ZYIVwvvPkW7azl/rOy+F3HYRZ2K5zeE9mmkhQppyv9sQFx0JM9UabnpPQ=="],
+    "json-schema-to-ts": ["json-schema-to-ts@3.1.1", "", { "dependencies": { "@babel/runtime": "^7.18.3", "ts-algebra": "^2.0.0" } }, "sha512-+DWg8jCJG2TEnpy7kOm/7/AxaYoaRbjVB4LFZLySZlWn8exGs3A4OLJR966cVvU26N7X9TWxl+Jsw7dzAqKT6g=="],
+    "standardwebhooks": ["standardwebhooks@1.0.0", "", { "dependencies": { "@stablelib/base64": "^1.0.0", "fast-sha256": "^1.3.0" } }, "sha512-BbHGOQK9olHPMvQNHWul6MYlrRTAOKn03rOe4A8O3CLWhNf4YHBqq2HJKKC+sfqpxiBY52pNeesD6jIiLDz8jg=="],
+    "ts-algebra": ["ts-algebra@2.0.0", "", {}, "sha512-FPAhNPFMrkwz76P7cdjdmiShwMynZYN6SgOujD1urY4oNm80Ou9oMdmbR45LotcKOXoy7wSmHkRFE6Mxbrhefw=="],
+  }
+}