@zelari/core 1.14.3 → 1.15.0

package/dist/agents/skills/builtin/schemaLoop.d.ts

@@ -0,0 +1,2 @@
+export {};
+//# sourceMappingURL=schemaLoop.d.ts.map

package/dist/agents/skills/builtin/schemaLoop.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"schemaLoop.d.ts","sourceRoot":"","sources":["../../../../src/agents/skills/builtin/schemaLoop.ts"],"names":[],"mappings":""}

package/dist/agents/skills/builtin/schemaLoop.js

@@ -0,0 +1,86 @@
+import { registerCodingSkill } from '../../skills.js';
+const schemaLoop = {
+    id: 'schema-loop',
+    version: '1.0.0',
+    name: 'Schema Loop (certify before done)',
+    description: 'Physicist-style coding harness: keep an explicit hypothesis, define certifiable checks, ' +
+        'run_backtest before claiming done, and treat prediction mismatches as plan-voiding surprises. ' +
+        'When ≥2 hypotheses remain, run one discriminating experiment first.',
+    category: 'debug',
+    requiredRoles: [],
+    requiredTools: [
+        'update_world_hypothesis',
+        'set_world_checks',
+        'run_backtest',
+        'record_world_observation',
+        'bash',
+        'read_file',
+    ],
+    estimatedCost: 'medium',
+    enabledByDefault: true,
+    builtin: true,
+    triggers: [
+        'Non-trivial bug or multi-step implementation',
+        'You are about to claim the task is done',
+        'Tests or typecheck failed after edits',
+        'Two or more competing root-cause hypotheses',
+        'Surprising tool result (wrong exit code, missing path, shell mismatch)',
+    ],
+    antiPatterns: [
+        'Trivial one-line typo fix with no uncertainty',
+        'Pure documentation rewrite with no behavior change',
+        'User explicitly asked for a draft-only answer without running tools',
+    ],
+    requires: [],
+    relatedSkills: ['debug-with-rag', 'root-cause-five-whys', 'reproduce-bug'],
+    tags: ['schema', 'backtest', 'hypothesis', 'verify', 'world-model', 'debug'],
+    examples: [
+        {
+            input: 'Fix the flaky headless test timeout',
+            output: {
+                hypothesis: 'Test hangs because registerMcpTools spawns real MCP servers; default 5s timeout is too low.',
+                checks: [
+                    { id: 'unit', command: 'npx vitest run tests/unit/cli-useChatTurn.test.ts' },
+                ],
+                next: 'set_world_checks → implement isolation → run_backtest → only then claim done',
+            },
+        },
+    ],
+    outputSchema: '{ hypothesis: string; checks: Array<{ id: string; command: string }>; backtestOk: boolean; surprises?: string[] }',
+    systemPromptFragment: `You are running the **Schema Loop** adapted for software engineering
+(inspired by Schema harness / ARC world-models — but for repos, not grids).
+
+## Core idea
+The latent model of the task is **explicit files under .zelari/world/**, not only chat context:
+- hypothesis.md — what you believe (objects = modules, mechanism = bug cause / design)
+- checks.json — certifiable shell checks (the "reality" oracle)
+- timeline.jsonl — append-only observations (do not rewrite history)
+
+## Outer loop
+1. **observe** — read errors, failing tests, tool results
+2. **deliberate** — update_world_hypothesis; set_world_checks
+3. **execute** — edit/run tools to change the repo
+4. **record** — record_world_observation on surprises; run_backtest to certify
+
+## Hard rules (reality outranks the model)
+1. **Never claim done** while the latest run_backtest has ok=false or total=0 without checks.
+2. After any batch of write/edit, prefer run_backtest (or set checks then backtest) before asserting success.
+3. **Surprise = void plan**: if a command exits unexpectedly, shell rejects POSIX on Windows, path missing, or tests fail differently than predicted — call record_world_observation (kind=surprise), revise hypothesis, do not keep patching blindly.
+4. **Discovery actions**: when ≥2 hypotheses remain consistent with evidence, design ONE discriminating experiment (read/grep/test) whose outcomes diverge; do not spray random edits.
+5. Prefer fast checks (single-file vitest, typecheck) over full suite until the model is stable.
+
+## Predict-then-act (soft)
+Before a risky bash/edit, state a short prediction in hypothesis or observation
+(expect exit 0, files touched). After the tool, if reality mismatches, log surprise.
+
+## Tools
+- update_world_hypothesis — write/append theory
+- set_world_checks — define certifiable commands
+- run_backtest — replay all checks; report exact pass/fail
+- record_world_observation — timeline ground truth
+
+Kill switch for tools: ZELARI_SCHEMA_LOOP=0.
+${''}`,
+};
+registerCodingSkill(schemaLoop);
+//# sourceMappingURL=schemaLoop.js.map

package/dist/agents/skills/builtin/schemaLoop.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"schemaLoop.js","sourceRoot":"","sources":["../../../../src/agents/skills/builtin/schemaLoop.ts"],"names":[],"mappings":"AAKA,OAAO,EAAE,mBAAmB,EAAE,MAAM,iBAAiB,CAAC;AAEtD,MAAM,UAAU,GAA0B;IACxC,EAAE,EAAE,aAAa;IACjB,OAAO,EAAE,OAAO;IAChB,IAAI,EAAE,mCAAmC;IACzC,WAAW,EACT,0FAA0F;QAC1F,gGAAgG;QAChG,qEAAqE;IACvE,QAAQ,EAAE,OAAO;IACjB,aAAa,EAAE,EAAE;IACjB,aAAa,EAAE;QACb,yBAAyB;QACzB,kBAAkB;QAClB,cAAc;QACd,0BAA0B;QAC1B,MAAM;QACN,WAAW;KACZ;IACD,aAAa,EAAE,QAAQ;IACvB,gBAAgB,EAAE,IAAI;IACtB,OAAO,EAAE,IAAI;IACb,QAAQ,EAAE;QACR,8CAA8C;QAC9C,yCAAyC;QACzC,uCAAuC;QACvC,6CAA6C;QAC7C,wEAAwE;KACzE;IACD,YAAY,EAAE;QACZ,+CAA+C;QAC/C,oDAAoD;QACpD,qEAAqE;KACtE;IACD,QAAQ,EAAE,EAAE;IACZ,aAAa,EAAE,CAAC,gBAAgB,EAAE,sBAAsB,EAAE,eAAe,CAAC;IAC1E,IAAI,EAAE,CAAC,QAAQ,EAAE,UAAU,EAAE,YAAY,EAAE,QAAQ,EAAE,aAAa,EAAE,OAAO,CAAC;IAC5E,QAAQ,EAAE;QACR;YACE,KAAK,EAAE,qCAAqC;YAC5C,MAAM,EAAE;gBACN,UAAU,EACR,6FAA6F;gBAC/F,MAAM,EAAE;oBACN,EAAE,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,mDAAmD,EAAE;iBAC7E;gBACD,IAAI,EAAE,8EAA8E;aACrF;SACF;KACF;IACD,YAAY,EACV,mHAAmH;IACrH,oBAAoB,EAAE;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAiCtB,EAAE,EAAE;CACL,CAAC;AAEF,mBAAmB,CAAC,UAAU,CAAC,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zelari/core",
-  "version": "1.14.3",
+  "version": "1.15.0",
   "description": "Zelari Code core library — provider-neutral agent loop (AgentHarness), multi-agent council orchestration, and built-in skills. Used by the zelari-code CLI; consumable by other agent frontends.",
   "author": "Anathema Studio <https://anathema-studio.com/>",
   "license": "MIT",
@@ -96,6 +96,10 @@
       "types": "./dist/agents/skills/builtin/testing.d.ts",
       "default": "./dist/agents/skills/builtin/testing.js"
     },
+    "./skills/builtin/schema-loop": {
+      "types": "./dist/agents/skills/builtin/schemaLoop.d.ts",
+      "default": "./dist/agents/skills/builtin/schemaLoop.js"
+    },
     "./events": {
       "types": "./dist/events/index.d.ts",
       "default": "./dist/events/index.js"