@hawkeyexl/manni 4.3.1 → 4.4.1

package/dist/chunk-RNVYGWIC.js

@@ -0,0 +1,30 @@
+#!/usr/bin/env node
+
+// src/shared/color.ts
+import pc from "picocolors";
+function palette(enabled) {
+  return pc.createColors(enabled);
+}
+function shouldColor(opts) {
+  const env = opts.env ?? process.env;
+  if (opts.noColor) return false;
+  if (env.NO_COLOR != null && env.NO_COLOR !== "") return false;
+  return Boolean(opts.isTTY);
+}
+function colorFor(command, isTTY, env) {
+  const noColor = colorOwner(command).opts().color === false;
+  return shouldColor({ noColor, isTTY, env });
+}
+function colorOwner(command) {
+  for (let c = command; c !== null; c = c.parent) {
+    if (c.options.some((o) => o.long === "--no-color")) return c;
+  }
+  return command;
+}
+
+export {
+  palette,
+  shouldColor,
+  colorFor
+};
+//# sourceMappingURL=chunk-RNVYGWIC.js.map

package/dist/chunk-RNVYGWIC.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/shared/color.ts"],"sourcesContent":["/**\n * Terminal color, decided once for every tool under the umbrella.\n *\n * clig.dev: color only on a TTY, never under `NO_COLOR` or `--no-color`. Each\n * tool's reporters take the decision as a boolean and build a palette from it,\n * so no reporter ever consults the environment on its own.\n */\nimport pc from \"picocolors\";\nimport type { Command } from \"commander\";\n\nexport type Colors = ReturnType<typeof pc.createColors>;\n\n/** Build a picocolors palette with color explicitly on or off. */\nexport function palette(enabled: boolean): Colors {\n  return pc.createColors(enabled);\n}\n\n/**\n * Decide whether to emit ANSI color: off when `--no-color`/`NO_COLOR`, on only\n * for a TTY otherwise. `forced` (from `--color`) overrides detection.\n */\nexport function shouldColor(opts: {\n  noColor?: boolean;\n  isTTY?: boolean;\n  env?: NodeJS.ProcessEnv;\n}): boolean {\n  const env = opts.env ?? process.env;\n  if (opts.noColor) return false;\n  if (env.NO_COLOR != null && env.NO_COLOR !== \"\") return false;\n  return Boolean(opts.isTTY);\n}\n\n/**\n * Whether `command`'s output gets colour: the domain's `--no-color` and\n * `NO_COLOR` turn it off, and otherwise only a TTY turns it on\n * (`shouldColor`). `isTTY` is passed uncoerced: Node leaves it undefined off\n * a terminal, never false, and `shouldColor` reads a missing one as \"not a\n * terminal\".\n */\nexport function colorFor(\n  command: Command,\n  isTTY: boolean | undefined,\n  env?: NodeJS.ProcessEnv,\n): boolean {\n  // commander maps --no-color to opts.color === false, on the command that\n  // declares it.\n  const noColor = colorOwner(command).opts().color === false;\n  return shouldColor({ noColor, isTTY, env });\n}\n\n/**\n * The nearest command, this one or an ancestor, that declares `--no-color`:\n * the domain's program, wherever it is mounted. Not the root: under the\n * umbrella that is `manni`, which has no `--no-color` of its own.\n */\nfunction colorOwner(command: Command): Command {\n  for (let c: Command | null = command; c !== null; c = c.parent) {\n    if (c.options.some((o) => o.long === \"--no-color\")) return c;\n  }\n  return command;\n}\n"],"mappings":";;;AAOA,OAAO,QAAQ;AAMR,SAAS,QAAQ,SAA0B;AAChD,SAAO,GAAG,aAAa,OAAO;AAChC;AAMO,SAAS,YAAY,MAIhB;AACV,QAAM,MAAM,KAAK,OAAO,QAAQ;AAChC,MAAI,KAAK,QAAS,QAAO;AACzB,MAAI,IAAI,YAAY,QAAQ,IAAI,aAAa,GAAI,QAAO;AACxD,SAAO,QAAQ,KAAK,KAAK;AAC3B;AASO,SAAS,SACd,SACA,OACA,KACS;AAGT,QAAM,UAAU,WAAW,OAAO,EAAE,KAAK,EAAE,UAAU;AACrD,SAAO,YAAY,EAAE,SAAS,OAAO,IAAI,CAAC;AAC5C;AAOA,SAAS,WAAW,SAA2B;AAC7C,WAAS,IAAoB,SAAS,MAAM,MAAM,IAAI,EAAE,QAAQ;AAC9D,QAAI,EAAE,QAAQ,KAAK,CAAC,MAAM,EAAE,SAAS,YAAY,EAAG,QAAO;AAAA,EAC7D;AACA,SAAO;AACT;","names":[]}

package/dist/chunk-SE4JN5XN.js

@@ -0,0 +1,19 @@
+#!/usr/bin/env node
+import {
+  ToolError
+} from "./chunk-G3T47CFV.js";
+
+// src/docevals/types.ts
+var DocevalsError = class extends ToolError {
+  constructor(message, code) {
+    super(message);
+    this.code = code;
+    this.name = "DocevalsError";
+  }
+  code;
+};
+
+export {
+  DocevalsError
+};
+//# sourceMappingURL=chunk-SE4JN5XN.js.map

package/dist/chunk-SE4JN5XN.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/docevals/types.ts"],"sourcesContent":["/**\n * Shared result model for manni docevals.\n *\n * Vocabulary follows the \"Docs as Evals\" methodology (Docs as Tests with AI):\n * every check is an *eval*; what differs is its *grader* (code-based, LLM-as-\n * judge, or human). Verdicts are binary pass/fail — aggregate rates emerge from\n * binary judgments, they are never assigned per item.\n */\nimport { ToolError } from \"../shared/errors.js\";\nimport type { Severity } from \"../shared/severity.js\";\n\n// The judge vocabulary is shared with the inference layer and is re-exported\n// here so manni docevals code and its consumers keep one import site. A second local\n// definition would be a second thing to keep in sync.\n//\n//   Match          — a single run's verdict; `partial` counts as fail\n//   Zone           — confidence-zone routing for AI-judged evals\n//   JudgeVerdict   — the structured verdict of one run (the book's shape)\n//   JudgeRun       — one run within an ensemble\n//   ConsensusResult — the aggregated outcome for one (page, eval) pair\nimport type { ConsensusResult } from \"@hawkeyexl/inference\";\n\nexport type {\n  ConsensusResult,\n  JudgeRun,\n  JudgeVerdict,\n  Match,\n  Zone,\n} from \"@hawkeyexl/inference\";\n\n/**\n * Regression evals guard behavior that must keep working (~100% target pass\n * rate). Capability evals probe what the docs/system can do (~70% target).\n */\nexport type EvalType = \"capability\" | \"regression\";\n\n/**\n * Finding severity for deterministically graded evals: the family scale,\n * `notice | warning | error`, from `src/shared/severity.ts`. Only `error`\n * affects exit codes.\n */\nexport type { Severity };\n\n/** How an eval is graded. `tool:regex` is the built-in `tool:*` kind; `registerGrader` adds others. */\nexport type GraderKind = \"ai\" | \"command\" | \"human\" | `tool:${string}`;\n\n/** A normalized finding from a deterministically graded eval (command or tool). */\nexport interface Finding {\n  /** Name of the eval that produced this finding. */\n  evalName: string;\n  /** Repo-relative path of the page the finding applies to. */\n  file: string;\n  /** Grader-specific rule id (e.g. \"regex/found\"), when available. */\n  ruleId?: string;\n  message: string;\n  severity: Severity;\n  /**\n   * This finding is about the *grader*, not the page: the tool could not be\n   * run, or its output could not be read. It fails the eval whatever severity\n   * the eval is configured at, because an eval configured `severity: warning`\n   * would otherwise pass while its check never executed (ADR 01022).\n   *\n   * A finding about the page carries the eval's severity as usual.\n   */\n  diagnostic?: boolean;\n  line?: number;\n  col?: number;\n}\n\n/** Result of one eval applied to one page. */\nexport interface EvalResult {\n  evalName: string;\n  /**\n   * Suite this result reports under. Stamped centrally by the engine, which\n   * already computes the mapping for the suite summaries — the alternative is\n   * threading it through every one of the ~30 places a result is constructed.\n   */\n  suite?: string;\n  type: EvalType;\n  grader: GraderKind;\n  file: string;\n  /**\n   * Where the page declares the eval: the item of its `evals` key that names\n   * it, in the page or in the manifest that supplied the key (proposal 0037).\n   * A suite eval the page never names points at its `eval-suite` key, or at\n   * the page alone when the suite is the config's default. `file` is spelled\n   * like `file`, and `line` is 1-based and absent when nothing records one.\n   *\n   * Stamped centrally by the engine, like `suite`, on every result it\n   * reports.\n   */\n  location?: { file: string; line?: number };\n  outcome: \"pass\" | \"fail\" | \"needs-review\" | \"skipped\" | \"error\";\n  /** Present for ai-graded evals. */\n  consensus?: ConsensusResult;\n  /** Present for command/tool-graded evals that produced findings. */\n  findings?: Finding[];\n  /** True when this run generated the eval's check script. */\n  generated?: boolean;\n  /** Set when a persisted human review resolved a needs-review outcome. */\n  via?: \"human-review\";\n  /**\n   * Findings the baseline already recorded and this run therefore suppressed\n   * (ADR 01017). Present only on a run with a baseline in effect, so its\n   * absence means \"no baseline\", not \"nothing was forgiven\".\n   */\n  baselined?: number;\n  skipReason?: string;\n  /**\n   * The eval's weight in its suite's pass rate. Stamped centrally by the\n   * engine alongside `suite`, and for the same reason: threading it through\n   * the ~30 sites that construct a result would be thirty chances to forget.\n   */\n  weight?: number;\n  /**\n   * Set when the model that judged this eval also produced what it graded.\n   *\n   * `content` — the machines attributed for what the eval's `target` reads\n   * name this model (`provenance` for the body, the `fields` of\n   * `meta-provenance` for the frontmatter, both for `raw`): the judge wrote\n   * what it grades. `criterion` — a `meta-provenance` entry for this model\n   * lists this eval under `evals`: the judge wrote the question. Both bias a\n   * verdict without invalidating it, so this rides along with a real outcome\n   * rather than replacing it, and reaches every reporter instead of stderr.\n   */\n  selfPreference?: { axis: \"content\" | \"criterion\"; model: string };\n  durationMs: number;\n}\n\n/** Per-suite aggregate. Pass rates emerge from binary outcomes. */\nexport interface SuiteSummary {\n  suite: string;\n  total: number;\n  passed: number;\n  failed: number;\n  needsReview: number;\n  skipped: number;\n  errored: number;\n  /**\n   * Weighted share of the graded set that passed, 1 when nothing ran.\n   *\n   * The graded set is pass + fail + error, as it has always been —\n   * `needs-review` and `skipped` are in neither half, so a page awaiting review\n   * neither helps nor hurts. Each eval contributes its `weight` (1 unless it\n   * says otherwise), and an eval that belongs to a criterion contributes\n   * nothing on its own: the criterion contributes once, for the group.\n   */\n  passRate: number;\n  /** From config: ~1.0 for regression suites, ~0.7 for capability suites. */\n  targetPassRate: number;\n  meetsTarget: boolean;\n  /**\n   * Set when a selection filter (`--eval` / `--suite`) was active, meaning the\n   * run measured part of this suite. `meetsTarget` is then always false and the\n   * suite cannot fail the run either: it has numbers, but no verdict.\n   */\n  partial?: boolean;\n  /**\n   * Criteria scored in this suite. Present only when the suite declares any,\n   * so its absence means \"none\", never \"none passed\".\n   *\n   * `suspended` counts criteria whose members were not all graded in this run.\n   * They are out of the rate entirely rather than counted as failures — the\n   * same reasoning as `partial` one level down (ADR 01018): a group measured\n   * in part has numbers but no verdict.\n   */\n  criteria?: {\n    total: number;\n    passed: number;\n    failed: number;\n    suspended: number;\n  };\n}\n\n/** Full run output consumed by reporters. */\nexport interface RunReport {\n  pages: number;\n  evalResults: EvalResult[];\n  suites: SuiteSummary[];\n  /**\n   * Judge telemetry. Counts, not money: the dollar figure this block used to\n   * carry was only as good as a price table this repo does not own, and was\n   * absent entirely for `claude-cli` and self-hosted endpoints (ADR 01019).\n   */\n  usage: {\n    totalTokens: number;\n    cachedEvals: number;\n    judgedEvals: number;\n  };\n  /** Script generations performed during this run (paths written). */\n  generated: string[];\n  exitCode: 0 | 1;\n}\n\n/** Operational/usage error → exit code 2. */\nexport class DocevalsError extends ToolError {\n  constructor(\n    message: string,\n    /**\n     * Set where a caller acts on the kind of error rather than reporting it.\n     * `nothing-resolved`: no page this run would check resolves an eval.\n     */\n    readonly code?: \"nothing-resolved\",\n  ) {\n    super(message);\n    this.name = \"DocevalsError\";\n  }\n}\n"],"mappings":";;;;;;AAmMO,IAAM,gBAAN,cAA4B,UAAU;AAAA,EAC3C,YACE,SAKS,MACT;AACA,UAAM,OAAO;AAFJ;AAGT,SAAK,OAAO;AAAA,EACd;AAAA,EAJW;AAKb;","names":[]}