vigiles 2.0.0 → 2.2.0
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.
- package/README.md +182 -134
- package/dist/action-gate.d.ts +28 -0
- package/dist/action-gate.js +73 -0
- package/dist/cli.js +705 -230
- package/dist/community-skills.d.ts +22 -0
- package/dist/community-skills.js +86 -0
- package/dist/compile-generator.d.ts +48 -0
- package/dist/compile-generator.js +322 -0
- package/dist/compile.d.ts +4 -0
- package/dist/compile.js +239 -45
- package/dist/coverage.d.ts +56 -0
- package/dist/coverage.js +178 -0
- package/dist/doc-refs.d.ts +60 -0
- package/dist/doc-refs.js +246 -0
- package/dist/eval.d.ts +62 -0
- package/dist/eval.js +174 -0
- package/dist/evolve.js +60 -125
- package/dist/frontmatter.d.ts +75 -0
- package/dist/frontmatter.js +263 -0
- package/dist/generate-schema.d.ts +51 -0
- package/dist/generate-schema.js +125 -0
- package/dist/generate-types.js +39 -1
- package/dist/harness-test.d.ts +38 -0
- package/dist/harness-test.js +129 -0
- package/dist/hash.d.ts +8 -0
- package/dist/hash.js +16 -0
- package/dist/inline.d.ts +22 -4
- package/dist/inline.js +60 -13
- package/dist/integrity.d.ts +29 -0
- package/dist/integrity.js +44 -0
- package/dist/linters.d.ts +5 -3
- package/dist/linters.js +144 -4
- package/dist/mock-model.d.ts +31 -0
- package/dist/mock-model.js +189 -0
- package/dist/orphans.d.ts +52 -0
- package/dist/orphans.js +124 -0
- package/dist/proofs.js +22 -16
- package/dist/refs.d.ts +44 -0
- package/dist/refs.js +144 -0
- package/dist/session.d.ts +97 -0
- package/dist/session.js +306 -0
- package/dist/sidecar.d.ts +35 -0
- package/dist/sidecar.js +102 -0
- package/dist/skill-driver.d.ts +77 -0
- package/dist/skill-driver.js +76 -0
- package/dist/skill-runtime.d.ts +101 -0
- package/dist/skill-runtime.js +289 -0
- package/dist/skill-test.d.ts +47 -0
- package/dist/skill-test.js +77 -0
- package/dist/spec.d.ts +119 -13
- package/dist/spec.js +51 -4
- package/dist/symbols.d.ts +30 -0
- package/dist/symbols.js +142 -0
- package/dist/test-utils.d.ts +8 -0
- package/dist/test-utils.js +41 -0
- package/dist/types.d.ts +34 -8
- package/dist/types.js +19 -0
- package/dist/validate.js +5 -3
- package/package.json +26 -5
- package/.claude/settings.json +0 -46
- package/.claude/settings.local.json +0 -8
- package/.github/workflows/ci.yml +0 -81
- package/.prettierignore +0 -1
- package/.vigiles/generated.d.ts +0 -205
- package/CLAUDE.md +0 -95
- package/CLAUDE.md.spec.ts +0 -142
- package/CONTRIBUTING.md +0 -121
- package/dist/action.d.ts.map +0 -1
- package/dist/action.js.map +0 -1
- package/dist/cli.d.ts.map +0 -1
- package/dist/cli.js.map +0 -1
- package/dist/cli.test.d.ts +0 -2
- package/dist/cli.test.d.ts.map +0 -1
- package/dist/cli.test.js +0 -650
- package/dist/cli.test.js.map +0 -1
- package/dist/compile.d.ts.map +0 -1
- package/dist/compile.js.map +0 -1
- package/dist/evolve.d.ts.map +0 -1
- package/dist/evolve.js.map +0 -1
- package/dist/freshness.d.ts +0 -67
- package/dist/freshness.d.ts.map +0 -1
- package/dist/freshness.js +0 -244
- package/dist/freshness.js.map +0 -1
- package/dist/freshness.test.d.ts +0 -2
- package/dist/freshness.test.d.ts.map +0 -1
- package/dist/freshness.test.js +0 -356
- package/dist/freshness.test.js.map +0 -1
- package/dist/generate-types.d.ts.map +0 -1
- package/dist/generate-types.js.map +0 -1
- package/dist/inline.d.ts.map +0 -1
- package/dist/inline.js.map +0 -1
- package/dist/inline.test.d.ts +0 -5
- package/dist/inline.test.d.ts.map +0 -1
- package/dist/inline.test.js +0 -152
- package/dist/inline.test.js.map +0 -1
- package/dist/linters.d.ts.map +0 -1
- package/dist/linters.js.map +0 -1
- package/dist/proofs.d.ts.map +0 -1
- package/dist/proofs.js.map +0 -1
- package/dist/proofs.test.d.ts +0 -9
- package/dist/proofs.test.d.ts.map +0 -1
- package/dist/proofs.test.js +0 -952
- package/dist/proofs.test.js.map +0 -1
- package/dist/spec.d.ts.map +0 -1
- package/dist/spec.js.map +0 -1
- package/dist/spec.test.d.ts +0 -2
- package/dist/spec.test.d.ts.map +0 -1
- package/dist/spec.test.js +0 -1222
- package/dist/spec.test.js.map +0 -1
- package/dist/types.d.ts.map +0 -1
- package/dist/types.js.map +0 -1
- package/dist/validate.d.ts.map +0 -1
- package/dist/validate.js.map +0 -1
- package/dist/validate.test.d.ts +0 -2
- package/dist/validate.test.d.ts.map +0 -1
- package/dist/validate.test.js +0 -531
- package/dist/validate.test.js.map +0 -1
- package/docs/agent-setup.md +0 -85
- package/docs/agent-workflows.md +0 -103
- package/docs/comparison.md +0 -71
- package/docs/freshness.md +0 -124
- package/docs/inline-mode.md +0 -119
- package/docs/linter-support.md +0 -166
- package/docs/spec-format.md +0 -194
- package/eslint.config.mjs +0 -79
- package/examples/CLAUDE.md +0 -54
- package/examples/CLAUDE.md.spec.ts +0 -65
- package/examples/SKILL.md +0 -50
- package/examples/SKILL.md.spec.ts +0 -57
- package/fixtures/example-project/CLAUDE.md +0 -11
- package/fixtures/example-project/package.json +0 -9
- package/fixtures/example-project/src/index.ts +0 -3
- package/fixtures/example-project/src/utils.test.ts +0 -2
- package/fixtures/example-project/src/utils.ts +0 -3
- package/logo.png +0 -0
- package/research/adoption-strategy.md +0 -111
- package/research/agent-integration.md +0 -145
- package/research/ai-code-quality.md +0 -197
- package/research/code-search-for-agents.md +0 -313
- package/research/competitive-landscape.md +0 -163
- package/research/doc-freshness.md +0 -516
- package/research/executable-specs.md +0 -368
- package/research/feature-ideas.md +0 -464
- package/research/formal-proofs-for-agents.md +0 -338
- package/research/fp-for-agent-harness.md +0 -150
- package/research/fp-for-deterministic-ai.md +0 -131
- package/research/self-evolving-specs.md +0 -298
- package/schemas/claude-md-strict.yml +0 -18
- package/schemas/claude-md.yml +0 -6
- package/schemas/skill-strict.yml +0 -12
- package/schemas/skill.yml +0 -5
- package/skills/audit-feedback-loop/SKILL.md +0 -76
- package/skills/edit-spec/SKILL.md +0 -131
- package/skills/enforce-rules-format/SKILL.md +0 -71
- package/skills/generate-logo/SKILL.md +0 -102
- package/skills/generate-rule/SKILL.md +0 -90
- package/skills/linter-docs/clippy.md +0 -241
- package/skills/linter-docs/eslint.md +0 -384
- package/skills/linter-docs/pylint.md +0 -288
- package/skills/linter-docs/rubocop.md +0 -277
- package/skills/linter-docs/ruff.md +0 -187
- package/skills/linter-docs/stylelint.md +0 -247
- package/skills/migrate-to-spec/SKILL.md +0 -124
- package/skills/pr-to-lint-rule/SKILL.md +0 -97
- package/skills/strengthen/SKILL.md +0 -168
- package/src/action.ts +0 -214
- package/src/cli.test.ts +0 -914
- package/src/cli.ts +0 -1631
- package/src/compile.ts +0 -691
- package/src/evolve.ts +0 -781
- package/src/freshness.test.ts +0 -449
- package/src/freshness.ts +0 -299
- package/src/generate-types.ts +0 -448
- package/src/inline.test.ts +0 -206
- package/src/inline.ts +0 -164
- package/src/linters.ts +0 -739
- package/src/proofs.test.ts +0 -1314
- package/src/proofs.ts +0 -849
- package/src/spec.test.ts +0 -1471
- package/src/spec.ts +0 -427
- package/src/types.ts +0 -117
- package/src/validate.test.ts +0 -701
- package/src/validate.ts +0 -381
- package/tsconfig.json +0 -23
|
@@ -0,0 +1,125 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
/**
|
|
3
|
+
* vigiles generate-schema — emit a JSON Schema for `vigiles:` frontmatter.
|
|
4
|
+
*
|
|
5
|
+
* Sibling of `generate-types` (which emits a `.d.ts` for the spec.ts path).
|
|
6
|
+
* This emits `.vigiles/schema.json` so a YAML LSP can autocomplete rule
|
|
7
|
+
* names and squiggle typos in markdown frontmatter:
|
|
8
|
+
*
|
|
9
|
+
* ---
|
|
10
|
+
* # yaml-language-server: $schema=./.vigiles/schema.json
|
|
11
|
+
* vigiles:
|
|
12
|
+
* enforce:
|
|
13
|
+
* - rule: eslint/no-consolee # red squiggle in VS Code
|
|
14
|
+
* ---
|
|
15
|
+
*
|
|
16
|
+
* The `rule` enum is populated from the project's ACTUAL enabled linter
|
|
17
|
+
* rules — the same discovery `generate-types` runs — so the schema reflects
|
|
18
|
+
* what is really enforceable, not a static catalog.
|
|
19
|
+
*/
|
|
20
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
21
|
+
exports.generateSchema = generateSchema;
|
|
22
|
+
const node_fs_1 = require("node:fs");
|
|
23
|
+
const node_path_1 = require("node:path");
|
|
24
|
+
const glob_1 = require("glob");
|
|
25
|
+
const generate_types_js_1 = require("./generate-types.js");
|
|
26
|
+
/**
|
|
27
|
+
* Rule references for config-declared custom linters. Mirrors
|
|
28
|
+
* `checkLinterRule`'s rulesDir lookup (any file `<rule>.*` is a rule) so the
|
|
29
|
+
* enum matches what `vigiles audit` accepts for these linters.
|
|
30
|
+
*/
|
|
31
|
+
function customRuleRefs(basePath, linters) {
|
|
32
|
+
const refs = [];
|
|
33
|
+
for (const [name, cfg] of Object.entries(linters ?? {})) {
|
|
34
|
+
const dirs = Array.isArray(cfg.rulesDir) ? cfg.rulesDir : [cfg.rulesDir];
|
|
35
|
+
for (const dir of dirs) {
|
|
36
|
+
if (!dir)
|
|
37
|
+
continue;
|
|
38
|
+
const full = (0, node_path_1.resolve)(basePath, dir);
|
|
39
|
+
if (!(0, node_fs_1.existsSync)(full))
|
|
40
|
+
continue;
|
|
41
|
+
for (const f of (0, glob_1.globSync)("*.*", { cwd: full, nodir: true })) {
|
|
42
|
+
refs.push(`${name}/${f.replace(/\.[^.]+$/, "")}`);
|
|
43
|
+
}
|
|
44
|
+
}
|
|
45
|
+
}
|
|
46
|
+
return refs;
|
|
47
|
+
}
|
|
48
|
+
/**
|
|
49
|
+
* Build a JSON Schema for the `vigiles:` frontmatter block from the
|
|
50
|
+
* project's discovered linter rules. When no rules are discoverable the
|
|
51
|
+
* `rule` field falls back to a freeform string so the schema never
|
|
52
|
+
* false-flags a valid reference.
|
|
53
|
+
*/
|
|
54
|
+
function generateSchema(options = {}) {
|
|
55
|
+
const basePath = options.basePath ?? process.cwd();
|
|
56
|
+
const { linters } = (0, generate_types_js_1.generateTypes)({ basePath });
|
|
57
|
+
const ruleNames = new Set();
|
|
58
|
+
for (const { linter, rules } of linters) {
|
|
59
|
+
for (const rule of rules) {
|
|
60
|
+
ruleNames.add(`${linter}/${rule}`);
|
|
61
|
+
// ESLint also accepts scoped-plugin rules in bare form (the scope is
|
|
62
|
+
// treated as the linter, e.g. "@typescript-eslint/no-explicit-any").
|
|
63
|
+
// Emit both so neither form produces a false squiggle.
|
|
64
|
+
if (linter === "eslint" && rule.includes("/")) {
|
|
65
|
+
ruleNames.add(rule);
|
|
66
|
+
}
|
|
67
|
+
}
|
|
68
|
+
}
|
|
69
|
+
for (const ref of customRuleRefs(basePath, options.linters)) {
|
|
70
|
+
ruleNames.add(ref);
|
|
71
|
+
}
|
|
72
|
+
const sorted = [...ruleNames].sort();
|
|
73
|
+
const ruleSchema = sorted.length > 0 ? { enum: sorted } : { type: "string" };
|
|
74
|
+
const schema = {
|
|
75
|
+
$schema: "http://json-schema.org/draft-07/schema#",
|
|
76
|
+
$id: "https://vigiles.dev/frontmatter.schema.json",
|
|
77
|
+
title: "vigiles frontmatter",
|
|
78
|
+
description: "vigiles enforce rules declared in markdown YAML frontmatter (Level 1).",
|
|
79
|
+
type: "object",
|
|
80
|
+
properties: {
|
|
81
|
+
vigiles: {
|
|
82
|
+
type: "object",
|
|
83
|
+
additionalProperties: false,
|
|
84
|
+
properties: {
|
|
85
|
+
enforce: {
|
|
86
|
+
type: "array",
|
|
87
|
+
description: "Linter rules to enforce, verified by `vigiles audit`.",
|
|
88
|
+
items: {
|
|
89
|
+
type: "object",
|
|
90
|
+
additionalProperties: false,
|
|
91
|
+
required: ["rule", "why"],
|
|
92
|
+
properties: {
|
|
93
|
+
rule: {
|
|
94
|
+
description: "Linter rule reference, e.g. eslint/no-console. Verified to exist AND be enabled.",
|
|
95
|
+
...ruleSchema,
|
|
96
|
+
},
|
|
97
|
+
why: {
|
|
98
|
+
type: "string",
|
|
99
|
+
description: "Why this rule is enforced — shown to the agent as context.",
|
|
100
|
+
},
|
|
101
|
+
},
|
|
102
|
+
},
|
|
103
|
+
},
|
|
104
|
+
files: {
|
|
105
|
+
type: "array",
|
|
106
|
+
description: "File paths referenced by this instruction file, verified to exist by `vigiles audit`.",
|
|
107
|
+
items: { type: "string" },
|
|
108
|
+
},
|
|
109
|
+
commands: {
|
|
110
|
+
type: "array",
|
|
111
|
+
description: "Commands (npm scripts / script-runner invocations) referenced here, verified by `vigiles audit`.",
|
|
112
|
+
items: { type: "string" },
|
|
113
|
+
},
|
|
114
|
+
},
|
|
115
|
+
},
|
|
116
|
+
},
|
|
117
|
+
};
|
|
118
|
+
return {
|
|
119
|
+
schema,
|
|
120
|
+
json: JSON.stringify(schema, null, 2) + "\n",
|
|
121
|
+
ruleNames: sorted,
|
|
122
|
+
linters: linters.map((l) => ({ linter: l.linter, count: l.rules.length })),
|
|
123
|
+
};
|
|
124
|
+
}
|
|
125
|
+
//# sourceMappingURL=generate-schema.js.map
|
package/dist/generate-types.js
CHANGED
|
@@ -73,7 +73,7 @@ function discoverEslintRules(basePath) {
|
|
|
73
73
|
const rules = JSON.parse(output.trim() || "[]");
|
|
74
74
|
if (rules.length === 0)
|
|
75
75
|
return null;
|
|
76
|
-
return { linter: "eslint", rules, via: "flat config" };
|
|
76
|
+
return { linter: "eslint", rules, via: "flat config (v9+/v10)" };
|
|
77
77
|
}
|
|
78
78
|
catch {
|
|
79
79
|
return null;
|
|
@@ -199,6 +199,41 @@ function discoverRubocopRules(basePath) {
|
|
|
199
199
|
return null;
|
|
200
200
|
}
|
|
201
201
|
}
|
|
202
|
+
function discoverCedarPolicies(basePath) {
|
|
203
|
+
const ID_RE = /@id\("([^"]+)"\)/g;
|
|
204
|
+
const STATEMENT_RE = /\b(?:permit|forbid)\s*\(/;
|
|
205
|
+
const policies = new Set();
|
|
206
|
+
for (const dir of [".cedar", "cedar"]) {
|
|
207
|
+
const fullDir = (0, node_path_1.resolve)(basePath, dir);
|
|
208
|
+
if (!(0, node_fs_1.existsSync)(fullDir))
|
|
209
|
+
continue;
|
|
210
|
+
const files = (0, glob_1.globSync)("**/*.cedar", { cwd: fullDir, nodir: true });
|
|
211
|
+
for (const file of files) {
|
|
212
|
+
let content;
|
|
213
|
+
try {
|
|
214
|
+
content = (0, node_fs_1.readFileSync)((0, node_path_1.resolve)(fullDir, file), "utf-8");
|
|
215
|
+
}
|
|
216
|
+
catch {
|
|
217
|
+
continue;
|
|
218
|
+
}
|
|
219
|
+
const annotated = [...content.matchAll(ID_RE)].map((m) => m[1]);
|
|
220
|
+
if (annotated.length > 0) {
|
|
221
|
+
for (const id of annotated)
|
|
222
|
+
policies.add(id);
|
|
223
|
+
}
|
|
224
|
+
else if (STATEMENT_RE.test(content)) {
|
|
225
|
+
policies.add(file.replace(/\.cedar$/, "").replace(/\\/g, "/"));
|
|
226
|
+
}
|
|
227
|
+
}
|
|
228
|
+
}
|
|
229
|
+
if (policies.size === 0)
|
|
230
|
+
return null;
|
|
231
|
+
return {
|
|
232
|
+
linter: "cedar",
|
|
233
|
+
rules: [...policies].sort(),
|
|
234
|
+
via: ".cedar files",
|
|
235
|
+
};
|
|
236
|
+
}
|
|
202
237
|
function discoverClippyRules(basePath) {
|
|
203
238
|
try {
|
|
204
239
|
const cargoPath = (0, node_path_1.resolve)(basePath, "Cargo.toml");
|
|
@@ -300,6 +335,9 @@ function generateTypes(options = {}) {
|
|
|
300
335
|
const clippy = discoverClippyRules(basePath);
|
|
301
336
|
if (clippy)
|
|
302
337
|
linters.push(clippy);
|
|
338
|
+
const cedar = discoverCedarPolicies(basePath);
|
|
339
|
+
if (cedar)
|
|
340
|
+
linters.push(cedar);
|
|
303
341
|
const scripts = discoverNpmScripts(basePath);
|
|
304
342
|
const files = discoverProjectFiles(basePath, fileGlobs);
|
|
305
343
|
// Generate .d.ts
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
import { type ModelTurn } from "./mock-model.js";
|
|
2
|
+
export { scriptModel, type ModelTurn } from "./mock-model.js";
|
|
3
|
+
export interface HarnessTestSpec {
|
|
4
|
+
/** Fixture files to write in a fresh temp working dir (path → contents). */
|
|
5
|
+
readonly files?: Record<string, string>;
|
|
6
|
+
/** `.claude/settings.json` contents — the hooks/permissions under test. */
|
|
7
|
+
readonly settings?: unknown;
|
|
8
|
+
/** The scripted model turns the agent will take. */
|
|
9
|
+
readonly model: readonly ModelTurn[];
|
|
10
|
+
/** The user prompt. Default: "go". */
|
|
11
|
+
readonly prompt?: string;
|
|
12
|
+
/** Tools the agent may use. Default: Read Edit Write Bash. */
|
|
13
|
+
readonly allowedTools?: readonly string[];
|
|
14
|
+
/** Per-run wall-clock timeout in ms. Default 60000. */
|
|
15
|
+
readonly timeoutMs?: number;
|
|
16
|
+
}
|
|
17
|
+
export interface HarnessTestResult {
|
|
18
|
+
readonly exitCode: number;
|
|
19
|
+
readonly stdout: string;
|
|
20
|
+
/** Hook block messages and diagnostics land here. */
|
|
21
|
+
readonly stderr: string;
|
|
22
|
+
/** The temp working dir (inspect or clean it up). */
|
|
23
|
+
readonly cwd: string;
|
|
24
|
+
/** Number of model turns the agent took (mock turns served). */
|
|
25
|
+
readonly turns: number;
|
|
26
|
+
/** Final contents of a file under the working dir, or null if absent. */
|
|
27
|
+
file(path: string): string | null;
|
|
28
|
+
/** Remove the temp working dir. */
|
|
29
|
+
cleanup(): void;
|
|
30
|
+
}
|
|
31
|
+
/** Whether the `claude` CLI is available — harness tests need it. */
|
|
32
|
+
export declare function claudeAvailable(): boolean;
|
|
33
|
+
/**
|
|
34
|
+
* Run the real `claude` CLI against a scripted mock model, with the given
|
|
35
|
+
* fixture and settings (hooks). Deterministic — same script, same result.
|
|
36
|
+
*/
|
|
37
|
+
export declare function runHarnessTest(spec: HarnessTestSpec): Promise<HarnessTestResult>;
|
|
38
|
+
//# sourceMappingURL=harness-test.d.ts.map
|
|
@@ -0,0 +1,129 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.scriptModel = void 0;
|
|
4
|
+
exports.claudeAvailable = claudeAvailable;
|
|
5
|
+
exports.runHarnessTest = runHarnessTest;
|
|
6
|
+
/**
|
|
7
|
+
* vigiles — deterministic Claude Code harness testing.
|
|
8
|
+
*
|
|
9
|
+
* Test what your *harness* does — hooks, settings, skills, instruction files —
|
|
10
|
+
* without paying for or depending on a real model. `runHarnessTest` spins up the
|
|
11
|
+
* real `claude` CLI (so your real hooks and settings fire exactly as in
|
|
12
|
+
* production) but points it at a scripted mock model (`src/mock-model.ts`), so
|
|
13
|
+
* the agent's turns are fixed and the outcome is reproducible. No API key, no
|
|
14
|
+
* cost, CI-friendly.
|
|
15
|
+
*
|
|
16
|
+
* const r = await runHarnessTest({
|
|
17
|
+
* settings: { hooks: { Stop: [{ hooks: [{ type: "command",
|
|
18
|
+
* command: "test -f DONE || { echo 'not done' >&2; exit 2; }" }] }] } },
|
|
19
|
+
* model: scriptModel([
|
|
20
|
+
* { text: "I'm done" }, // tries to stop → blocked
|
|
21
|
+
* { tool: "Bash", input: { command: "touch DONE" } },
|
|
22
|
+
* { text: "now done" },
|
|
23
|
+
* ]),
|
|
24
|
+
* });
|
|
25
|
+
* assert(JSON.parse(r.stdout).num_turns > 1); // the Stop hook fired
|
|
26
|
+
*
|
|
27
|
+
* The "steps" are the scripted model turns — their real home is deterministic
|
|
28
|
+
* harness testing, not production enforcement.
|
|
29
|
+
*
|
|
30
|
+
* Note: the simple mock drives the Bash tool and Stop hooks reliably; the
|
|
31
|
+
* Edit/Write tools are gated in headless mode and don't fire via the mock —
|
|
32
|
+
* drive file actions through Bash, or use the real-model eval tier (`eval.ts`)
|
|
33
|
+
* for Edit/Write hooks.
|
|
34
|
+
*/
|
|
35
|
+
const node_child_process_1 = require("node:child_process");
|
|
36
|
+
const node_fs_1 = require("node:fs");
|
|
37
|
+
const node_os_1 = require("node:os");
|
|
38
|
+
const node_path_1 = require("node:path");
|
|
39
|
+
const mock_model_js_1 = require("./mock-model.js");
|
|
40
|
+
var mock_model_js_2 = require("./mock-model.js");
|
|
41
|
+
Object.defineProperty(exports, "scriptModel", { enumerable: true, get: function () { return mock_model_js_2.scriptModel; } });
|
|
42
|
+
/** Whether the `claude` CLI is available — harness tests need it. */
|
|
43
|
+
function claudeAvailable() {
|
|
44
|
+
try {
|
|
45
|
+
return (0, node_child_process_1.spawnSync)("claude", ["--version"], { stdio: "ignore" }).status === 0;
|
|
46
|
+
}
|
|
47
|
+
catch {
|
|
48
|
+
return false;
|
|
49
|
+
}
|
|
50
|
+
}
|
|
51
|
+
function writeFixture(cwd, spec) {
|
|
52
|
+
for (const [p, content] of Object.entries(spec.files ?? {})) {
|
|
53
|
+
const full = (0, node_path_1.resolve)(cwd, p);
|
|
54
|
+
(0, node_fs_1.mkdirSync)((0, node_path_1.dirname)(full), { recursive: true });
|
|
55
|
+
(0, node_fs_1.writeFileSync)(full, content);
|
|
56
|
+
}
|
|
57
|
+
if (spec.settings !== undefined) {
|
|
58
|
+
// `{cwd}` in any hook command is substituted with the working dir, so a
|
|
59
|
+
// hook can reference an absolute path inside it (hooks don't run with the
|
|
60
|
+
// project dir as cwd).
|
|
61
|
+
const json = JSON.stringify(spec.settings, null, 2).replaceAll("{cwd}", cwd);
|
|
62
|
+
(0, node_fs_1.writeFileSync)((0, node_path_1.join)(cwd, "settings.json"), json);
|
|
63
|
+
}
|
|
64
|
+
}
|
|
65
|
+
function spawnClaude(args, cwd, baseUrl, timeoutMs) {
|
|
66
|
+
return new Promise((resolvePromise) => {
|
|
67
|
+
const child = (0, node_child_process_1.spawn)("claude", args, {
|
|
68
|
+
cwd,
|
|
69
|
+
env: {
|
|
70
|
+
...process.env,
|
|
71
|
+
ANTHROPIC_BASE_URL: baseUrl,
|
|
72
|
+
// Any value works — the mock ignores auth; this avoids needing a real key.
|
|
73
|
+
ANTHROPIC_API_KEY: "sk-vigiles-mock",
|
|
74
|
+
},
|
|
75
|
+
stdio: ["ignore", "pipe", "pipe"],
|
|
76
|
+
});
|
|
77
|
+
let stdout = "";
|
|
78
|
+
let stderr = "";
|
|
79
|
+
child.stdout.on("data", (d) => (stdout += d.toString()));
|
|
80
|
+
child.stderr.on("data", (d) => (stderr += d.toString()));
|
|
81
|
+
const timer = setTimeout(() => child.kill("SIGKILL"), timeoutMs);
|
|
82
|
+
child.on("close", (code) => {
|
|
83
|
+
clearTimeout(timer);
|
|
84
|
+
resolvePromise({ code: code ?? 0, stdout, stderr });
|
|
85
|
+
});
|
|
86
|
+
});
|
|
87
|
+
}
|
|
88
|
+
/**
|
|
89
|
+
* Run the real `claude` CLI against a scripted mock model, with the given
|
|
90
|
+
* fixture and settings (hooks). Deterministic — same script, same result.
|
|
91
|
+
*/
|
|
92
|
+
async function runHarnessTest(spec) {
|
|
93
|
+
const cwd = (0, node_fs_1.mkdtempSync)((0, node_path_1.join)((0, node_os_1.tmpdir)(), "vigiles-harness-"));
|
|
94
|
+
writeFixture(cwd, spec);
|
|
95
|
+
const mock = await (0, mock_model_js_1.startMock)(spec.model);
|
|
96
|
+
try {
|
|
97
|
+
const tools = spec.allowedTools ?? ["Read", "Edit", "Write", "Bash"];
|
|
98
|
+
const args = [
|
|
99
|
+
"-p",
|
|
100
|
+
spec.prompt ?? "go",
|
|
101
|
+
"--output-format",
|
|
102
|
+
"json",
|
|
103
|
+
"--model",
|
|
104
|
+
"claude-sonnet-4-5",
|
|
105
|
+
...(spec.settings !== undefined ? ["--settings", "settings.json"] : []),
|
|
106
|
+
"--allowedTools",
|
|
107
|
+
...tools,
|
|
108
|
+
];
|
|
109
|
+
const out = await spawnClaude(args, cwd, mock.url, spec.timeoutMs ?? 60000);
|
|
110
|
+
return {
|
|
111
|
+
exitCode: out.code,
|
|
112
|
+
stdout: out.stdout,
|
|
113
|
+
stderr: out.stderr,
|
|
114
|
+
cwd,
|
|
115
|
+
turns: mock.count,
|
|
116
|
+
file: (p) => {
|
|
117
|
+
const f = (0, node_path_1.resolve)(cwd, p);
|
|
118
|
+
return (0, node_fs_1.existsSync)(f) ? (0, node_fs_1.readFileSync)(f, "utf-8") : null;
|
|
119
|
+
},
|
|
120
|
+
cleanup: () => {
|
|
121
|
+
(0, node_fs_1.rmSync)(cwd, { recursive: true, force: true });
|
|
122
|
+
},
|
|
123
|
+
};
|
|
124
|
+
}
|
|
125
|
+
finally {
|
|
126
|
+
mock.close();
|
|
127
|
+
}
|
|
128
|
+
}
|
|
129
|
+
//# sourceMappingURL=harness-test.js.map
|
package/dist/hash.d.ts
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
1
|
+
declare const __brand: unique symbol;
|
|
2
|
+
export type SHA256Hash = string & {
|
|
3
|
+
readonly [__brand]: "SHA256Hash";
|
|
4
|
+
};
|
|
5
|
+
export declare function sha256short(data: string | Buffer): SHA256Hash;
|
|
6
|
+
export declare function assertNever(x: never): never;
|
|
7
|
+
export {};
|
|
8
|
+
//# sourceMappingURL=hash.d.ts.map
|
package/dist/hash.js
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.sha256short = sha256short;
|
|
4
|
+
exports.assertNever = assertNever;
|
|
5
|
+
const node_crypto_1 = require("node:crypto");
|
|
6
|
+
const HASH_LENGTH = 16;
|
|
7
|
+
function sha256short(data) {
|
|
8
|
+
return (0, node_crypto_1.createHash)("sha256")
|
|
9
|
+
.update(data)
|
|
10
|
+
.digest("hex")
|
|
11
|
+
.slice(0, HASH_LENGTH);
|
|
12
|
+
}
|
|
13
|
+
function assertNever(x) {
|
|
14
|
+
throw new Error(`Unexpected value: ${String(x)}`);
|
|
15
|
+
}
|
|
16
|
+
//# sourceMappingURL=hash.js.map
|
package/dist/inline.d.ts
CHANGED
|
@@ -25,8 +25,24 @@ export interface InlineRule {
|
|
|
25
25
|
/** 1-based line number of the comment in the source file. */
|
|
26
26
|
line: number;
|
|
27
27
|
}
|
|
28
|
+
/** A `<!-- vigiles:file <path> -->` reference (verified to exist). */
|
|
29
|
+
export interface InlineFileRef {
|
|
30
|
+
/** Project-relative path to verify exists. */
|
|
31
|
+
path: string;
|
|
32
|
+
/** 1-based line number of the comment in the source file. */
|
|
33
|
+
line: number;
|
|
34
|
+
}
|
|
35
|
+
/** A `<!-- vigiles:cmd "<command>" -->` reference (npm scripts verified). */
|
|
36
|
+
export interface InlineCmdRef {
|
|
37
|
+
/** Command to verify (npm scripts checked against package.json). */
|
|
38
|
+
command: string;
|
|
39
|
+
/** 1-based line number of the comment in the source file. */
|
|
40
|
+
line: number;
|
|
41
|
+
}
|
|
28
42
|
export interface InlineParseResult {
|
|
29
43
|
rules: InlineRule[];
|
|
44
|
+
files: InlineFileRef[];
|
|
45
|
+
commands: InlineCmdRef[];
|
|
30
46
|
/** Lines that look like vigiles: markers but failed to parse. */
|
|
31
47
|
errors: {
|
|
32
48
|
line: number;
|
|
@@ -46,13 +62,15 @@ export interface InlineParseResult {
|
|
|
46
62
|
*/
|
|
47
63
|
export declare function parseInlineRules(content: string): InlineParseResult;
|
|
48
64
|
/**
|
|
49
|
-
* True if the content contains at least one parseable vigiles
|
|
50
|
-
* rule
|
|
51
|
-
* `require-spec` validation to
|
|
65
|
+
* True if the content contains at least one parseable vigiles inline marker —
|
|
66
|
+
* an `enforce` rule, a `file` reference, or a `cmd` reference (ignoring fenced
|
|
67
|
+
* code blocks and malformed markers). Used by `require-spec` validation to
|
|
68
|
+
* treat inline mode as spec-equivalent: a file that pins even a single path is
|
|
69
|
+
* meaningfully managed.
|
|
52
70
|
*
|
|
53
71
|
* Deliberately delegates to `parseInlineRules` so a loose prefix regex
|
|
54
72
|
* can't satisfy require-spec with a malformed marker that produces no
|
|
55
|
-
* real
|
|
73
|
+
* real reference.
|
|
56
74
|
*/
|
|
57
75
|
export declare function hasInlineRules(content: string): boolean;
|
|
58
76
|
//# sourceMappingURL=inline.d.ts.map
|
package/dist/inline.js
CHANGED
|
@@ -29,6 +29,17 @@ exports.hasInlineRules = hasInlineRules;
|
|
|
29
29
|
* embedded quotes, they can move to spec mode.
|
|
30
30
|
*/
|
|
31
31
|
const ENFORCE_RE = /<!--\s*vigiles:enforce\s+([@A-Za-z0-9_/:.-]+)\s+"([^"\n]*)"\s*-->/;
|
|
32
|
+
/**
|
|
33
|
+
* Match `<!-- vigiles:file <path> -->`. The path is a single whitespace-free
|
|
34
|
+
* token (project-relative); spaces in paths are vanishingly rare and would be
|
|
35
|
+
* ambiguous against the closing `-->`.
|
|
36
|
+
*/
|
|
37
|
+
const FILE_RE = /<!--\s*vigiles:file\s+(\S+)\s*-->/;
|
|
38
|
+
/**
|
|
39
|
+
* Match `<!-- vigiles:cmd "<command>" -->`. The command is quoted because
|
|
40
|
+
* commands contain spaces (e.g. `npm run build`).
|
|
41
|
+
*/
|
|
42
|
+
const CMD_RE = /<!--\s*vigiles:cmd\s+"([^"\n]*)"\s*-->/;
|
|
32
43
|
/**
|
|
33
44
|
* Detects any `<!-- vigiles:<kind> -->` comment (valid or not) so we can
|
|
34
45
|
* surface errors for typos and reserved-but-unrecognized kinds. Uses a
|
|
@@ -36,6 +47,16 @@ const ENFORCE_RE = /<!--\s*vigiles:enforce\s+([@A-Za-z0-9_/:.-]+)\s+"([^"\n]*)"\
|
|
|
36
47
|
* circuit the pattern.
|
|
37
48
|
*/
|
|
38
49
|
const MARKER_RE = /<!--\s*vigiles:([A-Za-z_-]+)[^]*?-->/;
|
|
50
|
+
// vigiles markers handled by other subsystems (skill runtime gates, opt-outs).
|
|
51
|
+
// The inline *rule* parser skips them rather than flagging them as unknown.
|
|
52
|
+
const KNOWN_NON_RULE_MARKERS = new Set([
|
|
53
|
+
"disable",
|
|
54
|
+
"ignore",
|
|
55
|
+
"ignore-file",
|
|
56
|
+
"gate", // skill step gate (src/skill-runtime.ts)
|
|
57
|
+
"result", // skill result gate
|
|
58
|
+
"symbol", // symbol reference mark (src/refs.ts)
|
|
59
|
+
]);
|
|
39
60
|
/**
|
|
40
61
|
* Parse inline vigiles rules out of a markdown file's contents.
|
|
41
62
|
* Does not touch the filesystem and does not verify the rules against
|
|
@@ -48,6 +69,8 @@ const MARKER_RE = /<!--\s*vigiles:([A-Za-z_-]+)[^]*?-->/;
|
|
|
48
69
|
*/
|
|
49
70
|
function parseInlineRules(content) {
|
|
50
71
|
const rules = [];
|
|
72
|
+
const files = [];
|
|
73
|
+
const commands = [];
|
|
51
74
|
const errors = [];
|
|
52
75
|
const lines = content.split("\n");
|
|
53
76
|
let fenceChar = null;
|
|
@@ -97,15 +120,24 @@ function parseInlineRules(content) {
|
|
|
97
120
|
});
|
|
98
121
|
continue;
|
|
99
122
|
}
|
|
123
|
+
const fileMatch = FILE_RE.exec(scannable);
|
|
124
|
+
if (fileMatch) {
|
|
125
|
+
files.push({ path: fileMatch[1], line: i + 1 });
|
|
126
|
+
continue;
|
|
127
|
+
}
|
|
128
|
+
const cmdMatch = CMD_RE.exec(scannable);
|
|
129
|
+
if (cmdMatch) {
|
|
130
|
+
commands.push({ command: cmdMatch[1], line: i + 1 });
|
|
131
|
+
continue;
|
|
132
|
+
}
|
|
100
133
|
// Skip the compiled-file hash header (`<!-- vigiles:sha256:... -->`)
|
|
101
134
|
// entirely — it's not a rule marker and should not be reported.
|
|
102
135
|
if (/<!--\s*vigiles:sha\d+:/.test(scannable))
|
|
103
136
|
continue;
|
|
104
137
|
const markerMatch = MARKER_RE.exec(scannable);
|
|
105
138
|
if (markerMatch) {
|
|
106
|
-
// Looks like a vigiles marker but didn't parse
|
|
107
|
-
//
|
|
108
|
-
// unquoted why.
|
|
139
|
+
// Looks like a vigiles marker but didn't parse — surface it so users
|
|
140
|
+
// catch typos like "vigile:enforce" or an unquoted why/command.
|
|
109
141
|
const kind = markerMatch[1];
|
|
110
142
|
if (kind === "enforce") {
|
|
111
143
|
errors.push({
|
|
@@ -114,29 +146,44 @@ function parseInlineRules(content) {
|
|
|
114
146
|
raw: line.trim(),
|
|
115
147
|
});
|
|
116
148
|
}
|
|
117
|
-
else if (kind
|
|
118
|
-
|
|
119
|
-
|
|
149
|
+
else if (kind === "file") {
|
|
150
|
+
errors.push({
|
|
151
|
+
line: i + 1,
|
|
152
|
+
message: "Malformed vigiles:file — expected `<!-- vigiles:file <path> -->`",
|
|
153
|
+
raw: line.trim(),
|
|
154
|
+
});
|
|
155
|
+
}
|
|
156
|
+
else if (kind === "cmd") {
|
|
157
|
+
errors.push({
|
|
158
|
+
line: i + 1,
|
|
159
|
+
message: 'Malformed vigiles:cmd — expected `<!-- vigiles:cmd "<command>" -->`',
|
|
160
|
+
raw: line.trim(),
|
|
161
|
+
});
|
|
162
|
+
}
|
|
163
|
+
else if (!KNOWN_NON_RULE_MARKERS.has(kind)) {
|
|
120
164
|
errors.push({
|
|
121
165
|
line: i + 1,
|
|
122
|
-
message: `Unknown vigiles marker "${kind}". Only \`vigiles:enforce\`
|
|
166
|
+
message: `Unknown vigiles marker "${kind}". Only \`vigiles:enforce\`, \`vigiles:file\`, and \`vigiles:cmd\` are supported.`,
|
|
123
167
|
raw: line.trim(),
|
|
124
168
|
});
|
|
125
169
|
}
|
|
126
170
|
}
|
|
127
171
|
}
|
|
128
|
-
return { rules, errors };
|
|
172
|
+
return { rules, files, commands, errors };
|
|
129
173
|
}
|
|
130
174
|
/**
|
|
131
|
-
* True if the content contains at least one parseable vigiles
|
|
132
|
-
* rule
|
|
133
|
-
* `require-spec` validation to
|
|
175
|
+
* True if the content contains at least one parseable vigiles inline marker —
|
|
176
|
+
* an `enforce` rule, a `file` reference, or a `cmd` reference (ignoring fenced
|
|
177
|
+
* code blocks and malformed markers). Used by `require-spec` validation to
|
|
178
|
+
* treat inline mode as spec-equivalent: a file that pins even a single path is
|
|
179
|
+
* meaningfully managed.
|
|
134
180
|
*
|
|
135
181
|
* Deliberately delegates to `parseInlineRules` so a loose prefix regex
|
|
136
182
|
* can't satisfy require-spec with a malformed marker that produces no
|
|
137
|
-
* real
|
|
183
|
+
* real reference.
|
|
138
184
|
*/
|
|
139
185
|
function hasInlineRules(content) {
|
|
140
|
-
|
|
186
|
+
const r = parseInlineRules(content);
|
|
187
|
+
return r.rules.length + r.files.length + r.commands.length > 0;
|
|
141
188
|
}
|
|
142
189
|
//# sourceMappingURL=inline.js.map
|
|
@@ -0,0 +1,29 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Integrity check: verify a compiled markdown file hasn't been hand-edited.
|
|
3
|
+
*
|
|
4
|
+
* Every compiled file has a SHA-256 hash in its first line:
|
|
5
|
+
* <!-- vigiles:sha256:<hash> compiled from CLAUDE.md.spec.ts -->
|
|
6
|
+
*
|
|
7
|
+
* If the body content's hash no longer matches what's recorded, someone
|
|
8
|
+
* edited the compiled output directly. The check is one-pass and ~free.
|
|
9
|
+
*
|
|
10
|
+
* This is the entire freshness story now: no input fingerprinting, no
|
|
11
|
+
* recompile diffing. Those responsibilities belong elsewhere:
|
|
12
|
+
*
|
|
13
|
+
* - Hand-edit detection → this module
|
|
14
|
+
* - "Did the spec change?" → guard() rules emitting compile hooks
|
|
15
|
+
* - "Are referenced linter rules / files / scripts still valid?"
|
|
16
|
+
* → enforce() / file() / cmd() catch this at compile time
|
|
17
|
+
* - "Are committed compiled files actually fresh?"
|
|
18
|
+
* → CI runs `vigiles compile` then `git diff --exit-code`
|
|
19
|
+
*/
|
|
20
|
+
export interface IntegrityResult {
|
|
21
|
+
intact: boolean;
|
|
22
|
+
reason?: string;
|
|
23
|
+
}
|
|
24
|
+
/**
|
|
25
|
+
* Check whether the compiled markdown's SHA-256 hash matches its body.
|
|
26
|
+
* Files without a hash header are treated as hand-written (intact).
|
|
27
|
+
*/
|
|
28
|
+
export declare function checkIntegrity(content: string): IntegrityResult;
|
|
29
|
+
//# sourceMappingURL=integrity.d.ts.map
|
|
@@ -0,0 +1,44 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
/**
|
|
3
|
+
* Integrity check: verify a compiled markdown file hasn't been hand-edited.
|
|
4
|
+
*
|
|
5
|
+
* Every compiled file has a SHA-256 hash in its first line:
|
|
6
|
+
* <!-- vigiles:sha256:<hash> compiled from CLAUDE.md.spec.ts -->
|
|
7
|
+
*
|
|
8
|
+
* If the body content's hash no longer matches what's recorded, someone
|
|
9
|
+
* edited the compiled output directly. The check is one-pass and ~free.
|
|
10
|
+
*
|
|
11
|
+
* This is the entire freshness story now: no input fingerprinting, no
|
|
12
|
+
* recompile diffing. Those responsibilities belong elsewhere:
|
|
13
|
+
*
|
|
14
|
+
* - Hand-edit detection → this module
|
|
15
|
+
* - "Did the spec change?" → guard() rules emitting compile hooks
|
|
16
|
+
* - "Are referenced linter rules / files / scripts still valid?"
|
|
17
|
+
* → enforce() / file() / cmd() catch this at compile time
|
|
18
|
+
* - "Are committed compiled files actually fresh?"
|
|
19
|
+
* → CI runs `vigiles compile` then `git diff --exit-code`
|
|
20
|
+
*/
|
|
21
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
22
|
+
exports.checkIntegrity = checkIntegrity;
|
|
23
|
+
const hash_js_1 = require("./hash.js");
|
|
24
|
+
const HASH_LINE_RE = /^<!-- vigiles:sha256:([a-f0-9]+) compiled from (.+) -->\r?\n\r?\n?/;
|
|
25
|
+
/**
|
|
26
|
+
* Check whether the compiled markdown's SHA-256 hash matches its body.
|
|
27
|
+
* Files without a hash header are treated as hand-written (intact).
|
|
28
|
+
*/
|
|
29
|
+
function checkIntegrity(content) {
|
|
30
|
+
const match = content.match(HASH_LINE_RE);
|
|
31
|
+
if (!match) {
|
|
32
|
+
return { intact: true, reason: "No hash header (hand-written file)" };
|
|
33
|
+
}
|
|
34
|
+
const expectedHash = match[1];
|
|
35
|
+
const body = content.replace(HASH_LINE_RE, "");
|
|
36
|
+
if ((0, hash_js_1.sha256short)(body) !== expectedHash) {
|
|
37
|
+
return {
|
|
38
|
+
intact: false,
|
|
39
|
+
reason: "Compiled file was modified directly — edit the .spec.ts source and recompile",
|
|
40
|
+
};
|
|
41
|
+
}
|
|
42
|
+
return { intact: true };
|
|
43
|
+
}
|
|
44
|
+
//# sourceMappingURL=integrity.js.map
|
package/dist/linters.d.ts
CHANGED
|
@@ -3,10 +3,11 @@
|
|
|
3
3
|
*
|
|
4
4
|
* Verifies that linter rule references (e.g., "eslint/no-console") point to
|
|
5
5
|
* real rules that exist and are enabled in project config. Supports:
|
|
6
|
-
* ESLint, Stylelint (Node API), Ruff, Clippy, Pylint, RuboCop (CLI)
|
|
6
|
+
* ESLint, Stylelint (Node API), Ruff, Clippy, Pylint, RuboCop (CLI),
|
|
7
|
+
* Cedar (filesystem policies for AWS Bedrock AgentCore / Vectimus).
|
|
7
8
|
*
|
|
8
|
-
* This is the core moat — no other tool resolves rules
|
|
9
|
-
* and checks config-enabled status.
|
|
9
|
+
* This is the core moat — no other tool resolves rules across 7 catalog APIs
|
|
10
|
+
* (6 linters + Cedar policy language) and checks config-enabled status.
|
|
10
11
|
*/
|
|
11
12
|
export type ConfigEnabledStatus = "enabled" | "disabled" | "unknown";
|
|
12
13
|
export interface LinterCheckResult {
|
|
@@ -23,6 +24,7 @@ export interface DetectedLinter {
|
|
|
23
24
|
}
|
|
24
25
|
/** @internal */ export declare function extractLinterName(enforcedBy: string): string;
|
|
25
26
|
/** @internal */ export declare function extractRuleName(enforcedBy: string): string | null;
|
|
27
|
+
/** @internal */ export declare function clearCedarCache(): void;
|
|
26
28
|
/**
|
|
27
29
|
* Check a single linter rule reference (e.g., "eslint/no-console").
|
|
28
30
|
*
|