@llm4ts/shell 0.16.1 → 0.17.0

package/flows/fixtures/epic-stories/conto-bonifico.md

@@ -47,7 +47,7 @@ The expected split of the demo epic against the internet-banking portal fixture
     {
       "id": "payments-contract",
       "title": "Payments domain contract and stateful fake routes",
-      "description": "Declare the Payments HttpApi in src/contracts/payments.ts: list saved beneficiaries (name, IBAN), create a transfer (from account id, beneficiary name, IBAN, amount in cents, description, execution date) returning a pending transfer with an id, confirm a transfer with a six-digit SCA code (any code confirms except 000000, which is refused as UnprocessableEntity), list transfers newest first with their state (pending, confirmed, refused), and get one transfer. Implement src/contracts/payments.fake.ts with an in-memory store per page session so a created then confirmed transfer appears in the list, a reset function, and the exported paymentsDomain. Run pnpm openapi and commit contracts/openapi/payments.json. Follow src/contracts/profile.ts and profile.fake.ts exactly. Add src/contracts/payments.test.ts covering the fake routes through the typed client: create then confirm then list, and the refused code 000000.",
+      "description": "Declare the Payments HttpApi in src/contracts/payments.ts: list saved beneficiaries (name, IBAN), create a transfer from a payload with exactly these fields: fromAccountId (an Accounts domain account id), beneficiary as either { beneficiaryId } for a saved one or { name, iban } typed by the customer, amountCents, description, executionDate (ISO date), returning a pending transfer with an id, confirm a transfer with a six-digit SCA code: TransferState is exactly \"pending\" | \"confirmed\" | \"refused\"; any six-digit code other than 000000 moves the transfer to confirmed, the code 000000 moves it to refused and returns the refused transfer (HTTP 200, not an error), and a code that is not six digits is UnprocessableEntity; list transfers newest first with their state, and get one transfer. Implement src/contracts/payments.fake.ts with an in-memory store per page session so a created then confirmed transfer appears in the list, a reset function, and the exported paymentsDomain. Run pnpm openapi and commit contracts/openapi/payments.json. Follow src/contracts/profile.ts and profile.fake.ts exactly. Add src/contracts/payments.test.ts covering the fake routes through the typed client: create then confirm then list, and the refused code 000000.",
       "dependsOn": [],
       "owned": [
         "src/contracts/payments.ts",
@@ -68,7 +68,8 @@ The expected split of the demo epic against the internet-banking portal fixture
         "client.payments.confirm({ params: { transferId }, payload: { code } })",
         "client.payments.list()",
         "client.payments.get({ params: { transferId } })",
-        "src/contracts/payments.test.ts"
+        "src/contracts/payments.test.ts",
+        "TransferState = \"pending\" | \"confirmed\" | \"refused\""
       ]
     },
     {
@@ -136,10 +137,11 @@ The expected split of the demo epic against the internet-banking portal fixture
     {
       "id": "bonifico-form",
       "title": "Bonifico: new transfer with review, SCA confirmation and outcome",
-      "description": "Add the feature src/features/bonifico/nuovo/ with messages.ts (English and Italian), route.tsx exporting bonificoNuovoFeature (id bonifico), and BonificoScreen.tsx: a form (from account, beneficiary picked from saved beneficiaries or typed with IbanField, amount with parseEuro, description, execution date) validated by a pure function, a review step, a six-digit SCA code step calling client.payments.confirm, and an outcome panel for confirmed and refused, all through paymentsDomain and useAction. Add BonificoScreen.test.tsx in the house style: validation blocks a bad submit, a good transfer reaches the fake and is confirmed, code 000000 shows the refusal.",
+      "description": "Add the feature src/features/bonifico/nuovo/ with messages.ts (English and Italian), route.tsx exporting bonificoNuovoFeature (id bonifico), and BonificoScreen.tsx: a form (from account chosen among the customer's accounts loaded through accountsDomain from src/contracts/accounts.fake.ts, beneficiary picked from saved beneficiaries or typed with IbanField, amount with parseEuro, description, execution date) validated by a pure function, a review step, a six-digit SCA code step calling client.payments.confirm, and an outcome panel for confirmed and refused, all through paymentsDomain and useAction. Add BonificoScreen.test.tsx in the house style: validation blocks a bad submit, a good transfer reaches the fake and is confirmed, code 000000 shows the refusal.",
       "dependsOn": [
         "payments-contract",
-        "iban-field"
+        "iban-field",
+        "accounts-contract"
       ],
       "owned": [
         "src/features/bonifico/nuovo"

package/flows/modernize-pack-check.js

@@ -0,0 +1,166 @@
+// Legacy modernization phase -1: load a pack and match its rules against an estate; no LLM.
+//
+// The pack equivalent of the mock provider: a deterministic first success
+// before any paid phase. The pack is loaded exactly as survey and extract load
+// it, then the flow reports what it found and how each rule matches the
+// estate at `--repo`:
+//
+//   1. manifest — name, source, scaffold, specs/features dirs, gates, replay,
+//      equivalence policy, judge dimensions, prompt and reviewer sidecars.
+//   2. estate — the files `sources:` selects (minus `exclude:`), the files
+//      `programs:` selects, and for every `## Coverage:` and `## Survey:` rule
+//      the units it captures, with a sample of each so a regex typo is
+//      visible at a glance.
+//   3. verdict — hard failures (the pack does not load, `sources:` or
+//      `programs:` match nothing) abort with exit code 1; likely mistakes (a
+//      rule capturing no unit, a missing phase prompt, a scaffold path that
+//      does not exist, no reviewer lens, no judge dimension) are listed as
+//      warnings and the flow exits 0.
+//
+// Run: LLM4TS_PACK=packs/my-pack modernize-pack-check --repo ~/estates/legacy
+import { existsSync } from "node:fs";
+import { join } from "node:path";
+import * as Effect from "effect/Effect";
+import { FlowAborted } from "@llm4ts/flow/FlowError";
+import { Info } from "@llm4ts/flow/FlowEvents";
+import { stage } from "@llm4ts/flow/PlanExecution";
+import { coverageUnits, matchingFiles } from "@llm4ts/flow/SpecChecks";
+import { legacySourceWorkspaceLimits, workspaceLimitsFromEnv } from "@llm4ts/flow/Workspace";
+import { mock } from "@llm4ts/runner/Connectors";
+import { resolveFlowInput } from "@llm4ts/runner/FlowArgs";
+import { runFlowMain, runNode } from "@llm4ts/runner/FlowRunner";
+import { makeNodeWorkspace } from "@llm4ts/runner/NodeWorkspace";
+import { openPack } from "@llm4ts/runner/Packs";
+/** Prompt sidecars the modernization phases read, with the phase that reads each. */
+const phasePrompts = [
+    ["analysis", "extract"],
+    ["spec", "extract"],
+    ["bdd", "extract"],
+    ["plan", "extract"],
+    ["implement", "implement"],
+    ["review", "review"],
+    ["vectors", "verify"]
+];
+const sampleSize = 5;
+const sample = (units) => units.length === 0
+    ? "(none)"
+    : units.slice(0, sampleSize).join(", ") +
+        (units.length > sampleSize ? ` … +${units.length - sampleSize} more` : "");
+const plural = (count, noun, many = `${noun}s`) => `${count} ${count === 1 ? noun : many}`;
+const program = Effect.gen(function* () {
+    const input = yield* resolveFlowInput("Check the pack against the estate");
+    yield* runNode({
+        workDir: input.workDir,
+        workspace: input.workspace,
+        userPrompt: input.prompt,
+        // This phase makes no model call; the mock seat satisfies the one
+        // context shape the runner composes for every flow.
+        coder: mock,
+        environment: process.env
+    }, (context) => Effect.gen(function* () {
+        const say = (message) => context.events.publish(Info.make({ message }));
+        const warnings = [];
+        const failures = [];
+        const opened = yield* stage(context.events, "pack", openPack({
+            environment: process.env,
+            launchDir: input.workspace,
+            flowDir: import.meta.dirname
+        }));
+        const pack = opened.pack;
+        const packRoot = join(opened.workspace.root, opened.dir);
+        yield* stage(context.events, "manifest", Effect.gen(function* () {
+            yield* say(`pack '${pack.name}' (source: ${pack.source}) at ${packRoot}`);
+            yield* say(`specs-dir: ${pack.specsDir} · features-dir: ${pack.featuresDir}`);
+            if (pack.scaffold === undefined) {
+                yield* say("scaffold: (none — seed needs a non-empty target)");
+            }
+            else {
+                const scaffoldRoot = join(packRoot, pack.scaffold);
+                yield* say(`scaffold: ${pack.scaffold}`);
+                if (!existsSync(scaffoldRoot)) {
+                    warnings.push(`scaffold '${pack.scaffold}' does not exist at ${scaffoldRoot}`);
+                }
+            }
+            const gates = Object.entries(pack.gates);
+            yield* say(gates.length === 0
+                ? "gates: (none)"
+                : `gates: ${gates.map(([name, command]) => `${name} → ${command.join(" ")}`).join(" · ")}`);
+            if (gates.length === 0) {
+                warnings.push("## Gates is empty — implement and verify have nothing to enforce");
+            }
+            yield* say(pack.replay === undefined
+                ? "replay: (none — verify cannot run equivalence vectors)"
+                : `replay: ${pack.replay.join(" ")} (${pack.equivalence.ordering.toLowerCase()} comparison)`);
+            yield* say(`judge: ${plural(pack.judgeDimensions.length, "dimension")}` +
+                (pack.judgeDimensions.length === 0
+                    ? ""
+                    : ` — ${pack.judgeDimensions.map((d) => `${d.name} (0..${d.maxScore})`).join(", ")}`));
+            if (pack.judgeDimensions.length === 0) {
+                warnings.push("## Judge has no dimensions — extract's gate cannot score a spec");
+            }
+            const present = phasePrompts.filter(([name]) => pack.prompt(name) !== undefined);
+            const missing = phasePrompts.filter(([name]) => pack.prompt(name) === undefined);
+            yield* say(`prompts: ${present.length}/${phasePrompts.length} phase sidecars` +
+                (present.length === 0 ? "" : ` — ${present.map(([name]) => name).join(", ")}`));
+            for (const [name, phase] of missing) {
+                warnings.push(`prompts/${name}.md not found (read by ${phase})`);
+            }
+            yield* say(`reviewers: ${plural(pack.lenses.length, "lens", "lenses")}` +
+                (pack.lenses.length === 0 ? "" : ` — ${pack.lenses.map((l) => l.name).join(", ")}`));
+            if (pack.lenses.length === 0) {
+                warnings.push("no reviewers/*.md sidecar — review runs without pack lenses");
+            }
+            yield* say(`lessons: ${pack.lessons === undefined ? "(none yet)" : "present"}`);
+        }));
+        yield* stage(context.events, "estate", Effect.gen(function* () {
+            const repo = yield* makeNodeWorkspace(input.workDir, workspaceLimitsFromEnv(process.env, legacySourceWorkspaceLimits));
+            const sourcesRegex = pack.sources ?? ".*";
+            const sources = yield* matchingFiles(repo, sourcesRegex, pack.exclude);
+            yield* say(`sources: ${plural(sources.length, "file")} match '${sourcesRegex}'` +
+                (pack.exclude === undefined ? "" : ` minus '${pack.exclude}'`) +
+                ` — ${sample(sources)}`);
+            if (sources.length === 0) {
+                failures.push(`sources '${sourcesRegex}' matched no file under ${input.workDir}`);
+            }
+            if (pack.programs === undefined) {
+                yield* say("programs: (not set — every source file is a program)");
+            }
+            else {
+                const programs = yield* matchingFiles(repo, pack.programs, pack.exclude);
+                yield* say(`programs: ${plural(programs.length, "file")} match '${pack.programs}' — ${sample(programs)}`);
+                if (programs.length === 0) {
+                    failures.push(`programs '${pack.programs}' matched no file under ${input.workDir}`);
+                }
+            }
+            const report = Effect.fn("modernize-pack-check.rules")(function* (kind, rules) {
+                if (rules.length === 0) {
+                    yield* say(`${kind}: (no rules)`);
+                    return;
+                }
+                const units = yield* coverageUnits(repo, rules);
+                for (const rule of rules) {
+                    const found = units[rule.name] ?? [];
+                    yield* say(`${kind} '${rule.name}': ${plural(found.length, "unit")} — ${sample(found)}`);
+                    if (found.length === 0) {
+                        warnings.push(`${kind} rule '${rule.name}' captured no unit: files '${rule.files}', unit '${rule.unit}'`);
+                    }
+                }
+            });
+            yield* report("coverage", pack.coverage);
+            yield* report("survey", pack.survey);
+        }));
+        yield* stage(context.events, "verdict", Effect.gen(function* () {
+            for (const warning of warnings) {
+                yield* say(`warning: ${warning}`);
+            }
+            if (failures.length > 0) {
+                return yield* FlowAborted.make({
+                    message: `pack '${pack.name}' failed the check: ${failures.join("; ")}`
+                });
+            }
+            yield* say(`pack '${pack.name}' check passed with ${plural(warnings.length, "warning")} — ` +
+                `next: LLM4TS_PACK=${process.env.LLM4TS_PACK ?? opened.dir} modernize-survey --repo ${input.workDir}`);
+        }));
+    }));
+});
+runFlowMain(program);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@llm4ts/shell",
-  "version": "0.16.1",
+  "version": "0.17.0",
   "description": "Interactive shell and CLI for llm4ts: flow discovery, run-a-flow, and view",
   "license": "MIT",
   "type": "module",
@@ -50,10 +50,10 @@
   "dependencies": {
     "@effect/platform-node": "4.0.0-beta.102",
     "@effect/platform-node-shared": "4.0.0-beta.102",
-    "@llm4ts/core": "0.16.1",
-    "@llm4ts/modernize": "0.16.1",
-    "@llm4ts/flow": "0.16.1",
-    "@llm4ts/runner": "0.16.1"
+    "@llm4ts/core": "0.17.0",
+    "@llm4ts/flow": "0.17.0",
+    "@llm4ts/runner": "0.17.0",
+    "@llm4ts/modernize": "0.17.0"
   },
   "peerDependencies": {
     "effect": "4.0.0-beta.102"