dsh-model-everything 0.1.0 → 0.2.0

package/README.en-US.md

@@ -2,6 +2,13 @@
 
 <p align="center"><strong>English</strong> · <a href="README.md">中文</a></p>
 
+<p align="center">
+  <a href="https://github.com/AmethystLuna/dsh-model-everything/actions/workflows/dsh-bundle.yml"><img src="https://github.com/AmethystLuna/dsh-model-everything/actions/workflows/dsh-bundle.yml/badge.svg" alt="CI"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT"></a>
+  <a href="https://nodejs.org/"><img src="https://img.shields.io/badge/node-%3E%3D20-blue.svg" alt="Node 20+"></a>
+  <a href="https://afdian.com/a/amethystluna"><img src="https://img.shields.io/badge/%E7%88%B1%E5%8F%91%E7%94%B5-Afdian-946CE6" alt="Sponsor on Afdian"></a>
+</p>
+
 Abstract what a design or a piece of code claims into a model and let an engine run it: findings carry citations and hashes, so they can be recomputed and compared. The conclusion comes from a run rather than from reasoning it out: thinking it through becomes running it.
 
 A native DeepSeek Harness plugin. It turns the assertions a design makes **about behaviour** into models that can be run, reviewed and exported:
@@ -141,6 +148,8 @@ The limit: it stores and compares **model and diagram source**; a rendered pictu
 | `model_design_review` | Design review: the dependency graph's shape, plus the contracts the design declares about itself - layering, responsibility boundaries, substitutability, interface width |
 | `report_diff` | Compares two reports: what a change added, removed or changed |
 
+`model_verify`, `model_datamodel_verify` and `model_design_review` also take `modelPath`: a JSON file in the workspace, in any formatting (indentation, a BOM and a trailing newline all parse). It is the alternative to an inline `model`; passing both is refused.
+
 ## Reading a report
 
 What a tool returns is a report, and the outcome is in `verdict`: read `verdict`, never `ok`. `ok` carries a different historical meaning in each family — most set it to `true` for a failing review, and a composition sets it to `false` when any error-severity finding exists — so it cannot be read as a verdict, and reading it as one is how a reader concludes the opposite of what the report says.
@@ -154,7 +163,7 @@ What a tool returns is a report, and the outcome is in `verdict`: read `verdict`
 | `schema` | The versioned report contract, for example `model-everything/verify/v1`. Branch on this field rather than on the shape of the payload |
 | `truncated` | The search hit its ceiling and stopped. **A truncated search is not a pass**, it is no conclusion; a property that could not be decided appears as `_UNDECIDED` |
 
-The body is what comes back by default. For the whole structured result, pass `reportView: "full"`.
+The body is what comes back by default. A truncated check can be expanded on its own: pass `check: "<id>"` and every finding of that check prints. For the whole structured result, pass `reportView: "full"` — compact JSON, no field lost.
 
 When `ok` is `false` the verdict is `fail` as well, but it means something else: the tool did not run. The reason is in `errorCode` - invalid input, an unsupported diagram family, or a declared `intent` this view cannot serve. Such a refusal starts its `verdictReason` with `the tool did not run:`.
 
@@ -234,8 +243,6 @@ How it works, in four sentences:
 
 **Default and switch**: on by default. That default was measured: it reaches the one request class the plugin otherwise never sees, and a control task was not over-triggered. The price is about 3.2k characters per session; turn it off in `cordis.patch.yml` (below) or with the Plugins-page switch, which is also how to run a comparison - the same request with the gate off and on.
 
-`gateContent` can be replaced wholesale to try different wording; `tests/gate` still requires it to name every shipped skill.
-
 ## Configuration
 
 The routing component (on by default) has three keys:
@@ -251,11 +258,11 @@ The main row has one key, `interaction` (what happens when a tool call is offere
 Write it into the profile's `cordis.patch.yml` by row id:
 
 ```yaml
-- insert:
-    - id: dsh-model-everything-gate
-      name: 'dsh-model-everything/gate'
-      config:
-        enabled: false        # turn the injection off
+# Override that row's config by id: the whole config is replaced, not merged.
+# Do not add `name` here - a row with a name inserts a new row instead of changing this one.
+- id: dsh-model-everything-gate
+  config:
+    enabled: false        # turn the injection off
 ```
 
 ## Boundaries
@@ -317,3 +324,9 @@ The published work behind the checks — and which checks have none — is in [d
 ## License and security
 
 MIT, see [LICENSE](LICENSE). Security policy in [SECURITY.md](SECURITY.md).
+
+## Sponsor
+
+This plugin is open source under the MIT license. If it saves you time, you can support development on [Afdian](https://afdian.com/a/amethystluna).
+
+Sponsorship is voluntary and changes neither the license nor the features.

package/README.md

@@ -2,6 +2,13 @@
 
 <p align="center"><a href="README.en-US.md">English</a> · <strong>中文</strong></p>
 
+<p align="center">
+  <a href="https://github.com/AmethystLuna/dsh-model-everything/actions/workflows/dsh-bundle.yml"><img src="https://github.com/AmethystLuna/dsh-model-everything/actions/workflows/dsh-bundle.yml/badge.svg" alt="CI"></a>
+  <a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License: MIT"></a>
+  <a href="https://nodejs.org/"><img src="https://img.shields.io/badge/node-%3E%3D20-blue.svg" alt="Node 20+"></a>
+  <a href="https://afdian.com/a/amethystluna"><img src="https://img.shields.io/badge/%E7%88%B1%E5%8F%91%E7%94%B5-Afdian-946CE6" alt="在爱发电上赞助"></a>
+</p>
+
 把设计与代码里的断言抽象成模型，交给引擎跑一遍：结论带出处与哈希，能复算、能比对。让结论来自一次运行，而不是一次通盘推理：把「想清楚」变成「跑一遍」。
 
 DeepSeek Harness 原生插件。把设计与代码里**关于行为的断言**变成可运行、可审查、可导出的模型：
@@ -141,6 +148,8 @@ DeepSeek Harness 原生插件。把设计与代码里**关于行为的断言**
 | `model_design_review` | 设计审阅：依赖图形状 + 设计自身声明的契约（分层、职责边界、可替换性、接口宽度） |
 | `report_diff` | 两份报告比对：新增/消失/变化的发现 |
 
+`model_verify`、`model_datamodel_verify`、`model_design_review` 另收 `modelPath`：从工作区里的 JSON 文件读模型，缩进、BOM、尾随换行都不影响。它与内联的 `model` 二选一，两个都给会被拒绝。
+
 ## 怎么读报告
 
 工具的产出是一份报告，结论只在 `verdict` 里：**看 `verdict`，不看 `ok`**。`ok` 在各族的含义不同——多数族在评审失败时仍是 `true`，组合族在出现 error 级发现时是 `false`——所以它不能当判定读；把它当判定读，会得出与报告相反的结论。
@@ -154,7 +163,7 @@ DeepSeek Harness 原生插件。把设计与代码里**关于行为的断言**
 | `schema` | 报告的版本化契约，例如 `model-everything/verify/v1`。要按字段分支，就分支在它上面 |
 | `truncated` | 搜索撞到上限就停了。**截断不等于通过**，是没结论；某个性质没判定出来时，报告里会出现 `_UNDECIDED` |
 
-默认返回正文。要完整 JSON，传 `reportView: "full"`。
+默认返回正文。被截断的检查可以单独展开：传 `check: "<id>"`，该检查的全部条目都会打印。要完整结果，传 `reportView: "full"`——紧凑 JSON，对象不丢字段。
 
 `ok` 为 `false` 时判定也是 `fail`，但含义不同：工具根本没跑。原因写在 `errorCode` 里——输入不合法、图族不支持，或声明的 `intent` 服务不了。这类拒绝的 `verdictReason` 以 `the tool did not run:` 开头。
 
@@ -234,8 +243,6 @@ npx -p @deepseek-ai/dsh dsh plugin --profile <profile> add dsh-model-everything
 
 **开关与默认**：默认开。这个默认值是量出来的——它补上的是插件唯一接不住的那类请求，而对照任务没有被过度触发。代价是每个会话约 3.2k 字符；要关掉就改 `cordis.patch.yml`（见下）或翻插件页的开关。想做对照（同一个请求，关一次开一次）也用它。
 
-`gateContent` 可以整段替换，用来试别的措辞；`tests/gate` 仍要求它点名每个已发布的技能。
-
 ## 配置
 
 注入组件（默认开）有三个键：
@@ -251,11 +258,11 @@ npx -p @deepseek-ai/dsh dsh plugin --profile <profile> add dsh-model-everything
 按 row id 写进 profile 的 `cordis.patch.yml`：
 
 ```yaml
-- insert:
-    - id: dsh-model-everything-gate
-      name: 'dsh-model-everything/gate'
-      config:
-        enabled: false        # 关掉注入
+# 按 id 覆盖那一行的 config：整段替换，不做深合并。
+# 这里不要写 name——写了 name 是「插入一行」，不是「改这一行」。
+- id: dsh-model-everything-gate
+  config:
+    enabled: false        # 关掉注入
 ```
 
 ## 边界
@@ -317,3 +324,9 @@ MODEL_EVERYTHING_SPIN=/path/to/spin MODEL_EVERYTHING_GCC=/path/to/gcc npm run te
 ## 许可证与安全
 
 MIT，见 [LICENSE](LICENSE)。安全策略见 [SECURITY.md](SECURITY.md)。
+
+## 赞助
+
+本插件以 MIT 许可证开源。如果它节省了你的时间，可以在[爱发电](https://afdian.com/a/amethystluna)赞助开发。
+
+赞助完全自愿，不影响许可证，也不改变任何功能。

package/lib/compose-tool.js

@@ -33,6 +33,10 @@ export const modelComposeTool = defineTool({
             type: 'integer',
             description: 'Maximum composite states to explore. Default 10000.',
         },
+        check: {
+            type: 'string',
+            description: 'Print one check in full when the digest truncated it. Pass the check id the CHECKS line lists (for example A9); the per-check budget no longer applies, and an id this report does not carry is answered with the ids that it does.',
+        },
         reportView: {
             type: 'string',
             enum: ['digest', 'full'],
@@ -45,7 +49,7 @@ export const modelComposeTool = defineTool({
             description: 'Composition verification report with C1/C2 findings and, when shared locations are declared, C3/C4/C5.',
         },
         render(args, value) {
-            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full' }) }];
+            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full', check: args.check }) }];
         },
     },
     timeoutMs: 10_000,

package/lib/data-engine.js

@@ -1132,7 +1132,7 @@ function dataNextSteps(checks) {
     // check looked at the model. `pass` does not say that, and neither does a list of the failures.
     const skipped = checks.filter((check) => check.status === 'skip').map((check) => check.name + ' (' + check.id + ')');
     if (skipped.length > 0) {
-        steps.push('These examined nothing, so their silence is not a result: ' + skipped.join(', ') + '. Declare what each one would need, or accept that the model is unverified on those points.');
+        steps.push('Not checked: ' + skipped.join(', ') + ' examined nothing — silence there is not a pass. Declare what each needs, or treat the model as unverified on those points.');
     }
     if (warningCodes.length > 0)
         steps.push('Review the warning findings (' + warningCodes.join(', ') + '): a nullable field or a missing index is a decision, not silence.');

package/lib/data-tool.js

@@ -1,5 +1,6 @@
 import { defineTool } from '@deepseek-ai/dsh-tools';
 import { runDataVerification, DATA_ENGINE_SCHEMA_VERSION } from './data-engine.js';
+import { oneOf } from './model-file.js';
 import { optionRefusal, requireArrayOption, requireObjectOption } from './option-shape.js';
 import { renderReportText } from './report-view.js';
 export const DATAMODEL_VERIFY_TOOL_NAME = 'model_datamodel_verify';
@@ -15,8 +16,11 @@ export const modelDataModelVerifyTool = defineTool({
     parameters: {
         model: {
             type: 'json',
-            required: true,
-            description: 'DataModelV1 data model to verify.',
+            description: 'DataModelV1 data model to verify. Pass this, or modelPath, not both.',
+        },
+        modelPath: {
+            type: 'string',
+            description: 'Path to a JSON file holding the model, instead of inlining it. Formatting does not matter: whitespace, a BOM and pretty printing all parse. Relative paths resolve against the process working directory, which is the session workspace.',
         },
         beforeModel: {
             type: 'json',
@@ -38,6 +42,10 @@ export const modelDataModelVerifyTool = defineTool({
             type: 'json',
             description: 'Optional backup/restore pairs for DA7 rollback symmetry.',
         },
+        check: {
+            type: 'string',
+            description: 'Print one check in full when the digest truncated it. Pass the check id the CHECKS line lists (for example A9); the per-check budget no longer applies, and an id this report does not carry is answered with the ids that it does.',
+        },
         reportView: {
             type: 'string',
             enum: ['digest', 'full'],
@@ -50,7 +58,7 @@ export const modelDataModelVerifyTool = defineTool({
             description: 'model-everything-datamodel verification report with summary and per-check findings.',
         },
         render(args, value) {
-            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full' }) }];
+            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full', check: args.check }) }];
         },
     },
     timeoutMs: 10_000,
@@ -63,7 +71,7 @@ export const modelDataModelVerifyTool = defineTool({
             requireArrayOption(args.copyPairs, 'copyPairs', 'DATA_OPTION_INVALID');
             requireArrayOption(args.migrationMappings, 'migrationMappings', 'DATA_OPTION_INVALID');
             requireArrayOption(args.backupPairs, 'backupPairs', 'DATA_OPTION_INVALID');
-            return runDataVerification(args.model, {
+            return runDataVerification(oneOf(args.model, args.modelPath, 'model', 'modelPath', 'DATA_OPTION_INVALID'), {
                 beforeModel: args.beforeModel,
                 fieldMapping: args.fieldMapping,
                 copyPairs: args.copyPairs,

package/lib/design-tool.js

@@ -1,5 +1,6 @@
 import { defineTool } from '@deepseek-ai/dsh-tools';
 import { REPORT_SCHEMAS } from './engine.js';
+import { oneOf } from './model-file.js';
 import { runDesignReview } from './design-engine.js';
 import { renderReportText } from './report-view.js';
 export const DESIGN_TOOL_NAME = 'model_design_review';
@@ -18,9 +19,16 @@ export const designTool = defineTool({
     parameters: {
         model: {
             type: 'json',
-            required: true,
             description: 'DesignModelV1: {schemaVersion: 1, architecture: {nodes: [ids], edges: [[from, to]]}, types?: [{name, stereotype, extends?, implements?, operations?, transitions?, forbids?, responsibilities?, dependsOn?, uses?, layer?}], declare?: {layers?, boundaries?, allowedEdges?, requiredEdges?}}. The schema is closed: an unknown key is refused rather than ignored. `types` is what makes the principle checks possible - with no `extends` there is no substitutability to decide, with no `uses` no interface can be called fat, and no `responsibilities` means nothing to compare the dependencies against.',
         },
+        modelPath: {
+            type: 'string',
+            description: 'Path to a JSON file holding the model, instead of inlining it. Formatting does not matter: whitespace, a BOM and pretty printing all parse. Relative paths resolve against the process working directory, which is the session workspace.',
+        },
+        check: {
+            type: 'string',
+            description: 'Print one check in full when the digest truncated it. Pass the check id the CHECKS line lists (for example A9); the per-check budget no longer applies, and an id this report does not carry is answered with the ids that it does.',
+        },
         reportView: {
             type: 'string',
             enum: ['digest', 'full'],
@@ -33,23 +41,23 @@ export const designTool = defineTool({
             description: 'Design review report: every check with its status, `violations` (they decide the verdict), `signals`, `facts` (graph features), summary and next steps.',
         },
         render(args, value) {
-            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full' }) }];
+            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full', check: args.check }) }];
         },
     },
     timeoutMs: 10_000,
     isConcurrencySafe: () => true,
     async execute(args) {
-        if (args.model === undefined) {
+        if (args.model === undefined && args.modelPath === undefined) {
             return {
                 ok: false,
                 ran: false,
                 verdict: 'fail',
-                verdictReason: 'the tool did not run: pass `model`, a DesignModelV1 with schemaVersion, architecture and the declarations to review it against',
+                verdictReason: 'the tool did not run: pass `model` (a DesignModelV1 with schemaVersion, architecture and the declarations) or `modelPath` (a JSON file holding one)',
                 schema: REPORT_SCHEMAS.design,
                 errorCode: 'DESIGN_INPUT',
-                error: 'design review needs `model`',
+                error: 'design review needs `model` or `modelPath`',
             };
         }
-        return runDesignReview(args.model);
+        return runDesignReview(oneOf(args.model, args.modelPath, 'model', 'modelPath', 'DESIGN_OPTION_INVALID'));
     },
 });

package/lib/engine.js

@@ -52,7 +52,7 @@ function verificationNextSteps(checks, coverage) {
     // named check looked at the model. A list of failures does not say that, and neither does `pass`.
     const skipped = checks.filter((check) => check.status === 'skip').map((check) => check.name + ' (' + check.id + ')');
     if (skipped.length > 0) {
-        steps.push('These examined nothing, so their silence is not a result: ' + skipped.join(', ') + '. Declare what each one would need, or accept that the model is unverified on those points.');
+        steps.push('Not checked: ' + skipped.join(', ') + ' examined nothing — silence there is not a pass. Declare what each needs, or treat the model as unverified on those points.');
     }
     // Named checks above, bare codes here. That is deliberate: each warning is a different kind of thing,
     // so there is no single name to give the set, and the sentence points at where each one IS explained

package/lib/model-file.js

@@ -0,0 +1,48 @@
+/**
+ * Read a JSON model from a file instead of requiring it inline.
+ *
+ * A model is the largest argument any of these tools takes, and it is the one that arrives as
+ * streamed tool-call JSON: the provider rejected one session with `tool input is invalid JSON`
+ * after the model spent sixteen steps hand-editing a nested model argument. Reading it from the
+ * workspace removes that failure mode and the tokens.
+ */
+import { readFileSync } from 'node:fs';
+import { resolve } from 'node:path';
+import { OptionShapeError } from './option-shape.js';
+/** Read and parse one JSON file. Formatting is irrelevant: whitespace, BOM and pretty printing all parse. */
+export function readJsonFile(path, option, code) {
+    if (typeof path !== 'string' || path.trim() === '') {
+        throw new OptionShapeError(code, `${option} must be a non-empty path to a JSON file.`);
+    }
+    const full = resolve(path);
+    let text;
+    try {
+        text = readFileSync(full, 'utf8');
+    }
+    catch (error) {
+        const reason = error instanceof Error ? error.message : String(error);
+        throw new OptionShapeError(code, `${option} could not be read: ${reason}. The path resolves against the process working directory; relative paths in a session resolve against the session's workspace.`);
+    }
+    try {
+        return JSON.parse(text.replace(/^\uFEFF/, ''));
+    }
+    catch (error) {
+        const reason = error instanceof Error ? error.message : String(error);
+        throw new OptionShapeError(code, `${option} is not valid JSON: ${reason}.`);
+    }
+}
+/**
+ * Exactly one of the inline value and the file path. Both is ambiguous, neither is a missing
+ * argument, and each gets its own sentence so the caller knows which one to drop.
+ */
+export function oneOf(value, path, inlineName, pathName, code) {
+    const hasValue = value !== undefined;
+    const hasPath = path !== undefined;
+    if (hasValue && hasPath) {
+        throw new OptionShapeError(code, `Pass either ${inlineName} or ${pathName}, not both: the file would be read and then ignored, or the other way round.`);
+    }
+    if (!hasValue && !hasPath) {
+        throw new OptionShapeError(code, `Pass ${inlineName} or ${pathName}: the model has to come from somewhere.`);
+    }
+    return hasPath ? readJsonFile(path, pathName, code) : value;
+}

package/lib/report-view.js

@@ -1,4 +1,4 @@
-const DEFAULTS = { maxFindingsPerCheck: 24, maxEvidenceChars: 160, includePassing: false, full: false };
+const DEFAULTS = { maxFindingsPerCheck: 24, maxEvidenceChars: 160, includePassing: false, full: false, check: undefined };
 const record = (value) => typeof value === 'object' && value !== null && !Array.isArray(value) ? value : undefined;
 const array = (value) => (Array.isArray(value) ? value : []);
 const text = (value) => (typeof value === 'string' ? value : undefined);
@@ -93,8 +93,20 @@ function findingLines(findings, options, seenFacts) {
         group.events.push(event);
         stateGroups.set(state, group);
     }
+    // The same event set on every state is one fact, not one per state: a 26-state machine whose
+    // states all ignore the same four events printed 24 near-identical lines, a quarter of the body.
+    // States that share a set are counted; a handful still name themselves, because there the names
+    // are what the reader acts on.
+    const byEventSet = new Map();
     for (const [state, group] of stateGroups) {
-        lines.push(`  ${state} (${String(group.events.length)}) [${group.severity}/${group.code}]: ${group.events.join(', ')}`);
+        const key = `${group.severity}|${group.code}|${group.events.join(',')}`;
+        const bucket = byEventSet.get(key) ?? { severity: group.severity, code: group.code, events: group.events, states: [] };
+        bucket.states.push(state);
+        byEventSet.set(key, bucket);
+    }
+    for (const group of byEventSet.values()) {
+        const who = group.states.length > 3 ? `${String(group.states.length)} states` : group.states.join(', ');
+        lines.push(`  ${who} (${String(group.events.length)}) [${group.severity}/${group.code}]: ${group.events.join(', ')}`);
     }
     const rendered = collapse(singles.map((finding) => findingLine(finding, options)));
     const budget = options.maxFindingsPerCheck;
@@ -219,13 +231,24 @@ function renderChecksReport(report, options) {
         .map(([status, ids]) => `${String(status)} ${ids.join(' ')}`)
         .join(' | ');
     const lines = [...headerLines(report, options), `CHECKS  ${inventory}`];
+    const requested = options.check;
+    const ids = checks.map((check) => text(check.id)).filter((id) => id !== undefined);
+    const selected = requested === undefined ? checks : checks.filter((check) => text(check.id) === requested);
+    // A name that matches nothing is answered with the names that exist: silently printing the whole
+    // report, or an empty body, both read as "that check was clean".
+    if (requested !== undefined && selected.length === 0) {
+        lines.push(`NOTE  no check \`${requested}\` in this report; it carries: ${ids.join(' ')}`);
+    }
+    // Asking for one check is a request for its lines: the per-check budget that keeps the digest
+    // small does not apply, since the caller has already paid for the narrowing.
+    const view = requested === undefined ? options : { ...options, includePassing: true, maxFindingsPerCheck: Number.MAX_SAFE_INTEGER };
     const seenFacts = new Set();
-    for (const check of checks) {
+    for (const check of selected) {
         const status = text(check.status) ?? 'skip';
         const findings = array(check.findings).map((finding) => toViewFinding(finding, text(check.id)));
         if (status === 'pass' && !options.includePassing)
             continue;
-        const body = findingLines(findings, options, seenFacts);
+        const body = findingLines(findings, view, seenFacts);
         if (status !== 'fail' && body.length === 0)
             continue;
         lines.push('');
@@ -380,8 +403,10 @@ function renderFallbackReport(report, options) {
  */
 export function renderReportText(value, options = {}) {
     const resolved = { ...DEFAULTS, ...options };
+    // Compact rather than pretty: `full` exists so a caller can hand the object on (report_diff matches
+    // on check + code + evidence), and indentation is a third of a large report's bytes for no reader.
     if (resolved.full)
-        return JSON.stringify(value, null, 2);
+        return JSON.stringify(value);
     const report = record(value);
     if (report === undefined)
         return JSON.stringify(value) ?? String(value);

package/lib/tool.js

@@ -2,6 +2,7 @@ import { defineTool } from '@deepseek-ai/dsh-tools';
 import { runVerification } from './engine.js';
 import { renderReportText } from './report-view.js';
 import { OptionShapeError, optionRefusal, requireObjectOption } from './option-shape.js';
+import { oneOf } from './model-file.js';
 export const MODEL_VERIFY_TOOL_NAME = 'model_verify';
 /**
  * Model-visible DSH tool wrapping the bundled TypeScript verification engine.
@@ -18,8 +19,11 @@ export const modelVerifyTool = defineTool({
     parameters: {
         model: {
             type: 'json',
-            required: true,
-            description: 'LogicModelV1 state machine model to verify.',
+            description: 'LogicModelV1 state machine model to verify. Pass this, or modelPath, not both.',
+        },
+        modelPath: {
+            type: 'string',
+            description: 'Path to a JSON file holding the model, instead of inlining it. Formatting does not matter: whitespace, a BOM and pretty printing all parse. Relative paths resolve against the process working directory, which is the session workspace.',
         },
         maxStates: {
             type: 'integer',
@@ -37,6 +41,10 @@ export const modelVerifyTool = defineTool({
             type: 'json',
             description: 'Optional object mapping BEFORE state ids to AFTER state ids. Omit for identity mapping (same state names).',
         },
+        check: {
+            type: 'string',
+            description: 'Print one check in full when the digest truncated it. Pass the check id the CHECKS line lists (for example A9); the per-check budget no longer applies, and an id this report does not carry is answered with the ids that it does.',
+        },
         reportView: {
             type: 'string',
             enum: ['digest', 'full'],
@@ -49,7 +57,7 @@ export const modelVerifyTool = defineTool({
             description: 'this plugin verification report with summary and per-check findings.',
         },
         render(args, value) {
-            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full' }) }];
+            return [{ type: 'text', text: renderReportText(value, { full: args.reportView === 'full', check: args.check }) }];
         },
     },
     timeoutMs: 10_000,
@@ -71,7 +79,8 @@ export const modelVerifyTool = defineTool({
                     throw new OptionShapeError('VERIFY_OPTION_INVALID', 'stateMapping.' + from + ' is not a state id (it is ' + (Array.isArray(to) ? 'an array' : to === null ? 'null' : 'a ' + typeof to) + '). Every value is the AFTER state id the BEFORE state maps to, so it has to be a non-empty string.');
                 }
             }
-            return runVerification(args.model, {
+            const model = oneOf(args.model, args.modelPath, 'model', 'modelPath', 'VERIFY_OPTION_INVALID');
+            return runVerification(model, {
                 maxStates: args.maxStates,
                 maxPermutationEvents: args.maxPermutationEvents,
                 beforeModel: args.beforeModel,

package/lib/types/model-file.d.ts

@@ -0,0 +1,7 @@
+/** Read and parse one JSON file. Formatting is irrelevant: whitespace, BOM and pretty printing all parse. */
+export declare function readJsonFile(path: unknown, option: string, code: string): unknown;
+/**
+ * Exactly one of the inline value and the file path. Both is ambiguous, neither is a missing
+ * argument, and each gets its own sentence so the caller knows which one to drop.
+ */
+export declare function oneOf(value: unknown, path: unknown, inlineName: string, pathName: string, code: string): unknown;

package/lib/types/report-view.d.ts

@@ -12,6 +12,13 @@ export interface ReportViewOptions {
      * matches on check + code + evidence, so a body cannot stand in for the object.
      */
     full?: boolean;
+    /**
+     * Print one check in full: every finding, evidence included, without the per-check budget the
+     * digest applies. The digest truncates a check's warnings and tells the reader to ask for it by
+     * id; before this option existed that sentence named something the tools could not do, and the
+     * only way to see the lines was `full`, which returns the whole structured report.
+     */
+    check?: string | undefined;
 }
 type Record_ = Record<string, unknown>;
 /** One normalized finding. `evidence` is kept because a finding's identity depends on it. */
@@ -22,7 +29,7 @@ export interface ViewFinding {
     evidence?: Record_;
     file?: string;
     line?: number;
-    check?: string;
+    check?: string | undefined;
 }
 /** Normalize one raw finding. Unknown fields are ignored; a missing message renders as the code. */
 export declare function toViewFinding(value: unknown, check?: string): ViewFinding;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-model-everything",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "description": "Modeling abstraction and automated review for designs and code, including a refactoring harness: UML diagrams, executable verification of machines and protocols with before/after regression, data-model and migration checks, concurrency claim routing, export to external checkers, dependency-graph review, and report diffing. Native plugin for DeepSeek Harness (dsh).",
   "type": "module",
   "main": "lib/index.js",
@@ -104,6 +104,10 @@
   },
   "license": "MIT",
   "repository": "https://github.com/AmethystLuna/dsh-model-everything",
+  "funding": {
+    "type": "custom",
+    "url": "https://afdian.com/a/amethystluna"
+  },
   "keywords": [
     "verification",
     "model-checking",