prehello 0.1.0 → 1.0.0

package/dist/diff.js

@@ -0,0 +1,182 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.DiffInputError = exports.DIFF_MARKER = void 0;
+exports.parseSnapshot = parseSnapshot;
+exports.diffSnapshots = diffSnapshots;
+exports.signedTok = signedTok;
+exports.headline = headline;
+exports.overThreshold = overThreshold;
+exports.renderDiffTerminal = renderDiffTerminal;
+exports.renderDiffMarkdown = renderDiffMarkdown;
+exports.diffJson = diffJson;
+exports.badgeJson = badgeJson;
+/**
+ * `prehello diff base.json head.json`: compare two saved reports (from `prehello --save` / `--json=`). Offline, no git:
+ * the GitHub Action (action.yml) produces the two snapshots. Item names come from the scanned repository, so every one
+ * goes through mdCode / sanitize before it is printed.
+ */
+const fmt_1 = require("./util/fmt");
+const markdown_1 = require("./render/markdown");
+const card_1 = require("./render/card");
+/** Starts with MD_MARKER_PREFIX (so --markdown=<file> can overwrite its own output) and is the sticky-comment marker the Action looks for. */
+exports.DIFF_MARKER = "<!-- generated by prehello diff -->";
+class DiffInputError extends Error {
+}
+exports.DiffInputError = DiffInputError;
+function parseSnapshot(text, what) {
+    let j;
+    try {
+        j = JSON.parse(text.replace(/^\uFEFF/, ""));
+    }
+    catch {
+        throw new DiffInputError(`${what} is not JSON; create it with: prehello --save ${what}`);
+    }
+    const s = j;
+    if (!s || typeof s !== "object" || s.tool?.name !== "prehello" || !Array.isArray(s.agents) || s.schemaVersion === undefined) {
+        throw new DiffInputError(`${what} is not a prehello report; create it with: prehello --save ${what}`);
+    }
+    return s;
+}
+const MAX_ITEMS = 10;
+const itemKey = (i) => `${i.kind}|${i.loadMode}|${i.path ?? i.label ?? i.id ?? ""}`;
+/** Counted when installed or when the repository has files for it (a CI runner has no agent installed). */
+const active = (a) => !!a && (!!a.detected || (a.items?.length ?? 0) > 0);
+function diffSnapshots(base, head) {
+    const notes = [];
+    if (String(base.schemaVersion) !== String(head.schemaVersion))
+        notes.push(`report schema differs (base ${base.schemaVersion}, head ${head.schemaVersion}); numbers may not be comparable`);
+    if (base.scan?.tokenizer?.method !== head.scan?.tokenizer?.method)
+        notes.push("base and head were counted with different tokenizers");
+    if (base.tool.version !== head.tool.version)
+        notes.push(`made by different prehello versions (base ${(0, fmt_1.sanitize)(base.tool.version, 20)}, head ${(0, fmt_1.sanitize)(head.tool.version, 20)}); part of the change may be the estimator`);
+    const ids = [...new Set([...base.agents, ...head.agents].map((a) => a.id))];
+    const agents = [];
+    for (const id of ids) {
+        const b = base.agents.find((a) => a.id === id);
+        const h = head.agents.find((a) => a.id === id);
+        if (!active(b) && !active(h))
+            continue;
+        const bt = active(b) ? b.totals.always.est : null;
+        const ht = active(h) ? h.totals.always.est : null;
+        const delta = (ht ?? 0) - (bt ?? 0);
+        const deferredDelta = (active(h) ? h.totals.deferred?.est ?? 0 : 0) - (active(b) ? b.totals.deferred?.est ?? 0 : 0);
+        if (b?.profile?.id && h?.profile?.id && b.profile.id !== h.profile.id)
+            notes.push(`${(0, fmt_1.sanitize)(h.name, 40)}: estimate profile changed (${(0, fmt_1.sanitize)(b.profile.id, 30)} -> ${(0, fmt_1.sanitize)(h.profile.id, 30)})`);
+        const bi = new Map((active(b) ? b.items ?? [] : []).filter((i) => i.loadMode === "always").map((i) => [itemKey(i), i]));
+        const hi = new Map((active(h) ? h.items ?? [] : []).filter((i) => i.loadMode === "always").map((i) => [itemKey(i), i]));
+        const changes = [];
+        for (const k of new Set([...bi.keys(), ...hi.keys()])) {
+            const x = bi.get(k);
+            const y = hi.get(k);
+            const bv = x?.tokens?.est ?? null;
+            const hv = y?.tokens?.est ?? null;
+            const d = Math.round((hv ?? 0) - (bv ?? 0));
+            if (x && y && d === 0)
+                continue;
+            const it = (y ?? x);
+            changes.push({ key: k, label: it.path ?? it.label ?? it.kind, kind: it.kind, base: x ? bv : null, head: y ? hv : null, delta: d });
+        }
+        changes.sort((p, q) => Math.abs(q.delta) - Math.abs(p.delta) || p.label.localeCompare(q.label));
+        agents.push({
+            id,
+            name: (h ?? b).name,
+            base: bt,
+            head: ht,
+            delta: Math.round(delta),
+            deltaPct: bt && bt >= 100 ? Math.round((delta / bt) * 1000) / 10 : null,
+            deferredDelta: Math.round(deferredDelta),
+            lowerBound: !!(b?.alwaysIsLowerBound || h?.alwaysIsLowerBound),
+            changes: changes.slice(0, MAX_ITEMS),
+            moreChanges: Math.max(0, changes.length - MAX_ITEMS),
+        });
+    }
+    return { base, head, agents, notes };
+}
+function signedTok(n) {
+    if (Math.round(n) === 0)
+        return "±0";
+    return (n > 0 ? "+" : "−") + (0, fmt_1.fmtTok)(Math.abs(n));
+}
+function headline(a) {
+    if (a.base === null)
+        return `${a.name}: new, ~${(0, fmt_1.fmtTok)(a.head ?? 0)} tokens before hello`;
+    if (a.head === null)
+        return `${a.name}: no longer detected (was ~${(0, fmt_1.fmtTok)(a.base)} tokens before hello)`;
+    if (a.delta === 0)
+        return `${a.name}: no change, ~${(0, fmt_1.fmtTok)(a.head)} tokens before hello`;
+    return `${a.name}: ${signedTok(a.delta)} tokens before hello (~${(0, fmt_1.fmtTok)(a.base)} → ~${(0, fmt_1.fmtTok)(a.head)}${a.deltaPct !== null ? `, ${a.deltaPct > 0 ? "+" : ""}${a.deltaPct}%` : ""})`;
+}
+const cell = (n) => (n === null ? "-" : (0, fmt_1.fmtTok)(n));
+const chg = (c) => (c.base === null ? "added" : c.head === null ? "removed" : "changed");
+function overThreshold(d, max) {
+    return d.agents.filter((a) => a.delta > max);
+}
+function renderDiffTerminal(d) {
+    const out = ["prehello diff (estimates; always-on context before your first message)", ""];
+    if (!d.agents.length)
+        out.push("No agent context found in either report.");
+    for (const a of d.agents) {
+        out.push((0, fmt_1.sanitize)(headline(a), 200));
+        for (const c of a.changes)
+            out.push(`    ${signedTok(c.delta).padStart(7)}  ${chg(c).padEnd(7)}  ${(0, fmt_1.sanitize)(c.label, 80)}`);
+        if (a.moreChanges)
+            out.push(`    … ${a.moreChanges} more changed item(s)`);
+        if (a.deferredDelta)
+            out.push(`    on demand (skill bodies, deferred tools): ${signedTok(a.deferredDelta)}`);
+    }
+    for (const n of d.notes)
+        out.push(`note: ${n}`);
+    return out.join("\n") + "\n";
+}
+function renderDiffMarkdown(d, opts = {}) {
+    const over = opts.threshold !== undefined ? overThreshold(d, opts.threshold) : [];
+    const out = [exports.DIFF_MARKER, "### prehello: context before hello", ""];
+    const moved = d.agents.filter((a) => a.delta !== 0 || a.base === null || a.head === null);
+    if (!d.agents.length)
+        out.push("No agent context (CLAUDE.md, AGENTS.md, rules, skills, MCP) found in this repository.");
+    else if (!moved.length)
+        out.push(`No change in always-on context. ${d.agents.map((a) => `${(0, markdown_1.mdCode)(a.name)} ~${(0, fmt_1.fmtTok)(a.head ?? 0)} tokens`).join(", ")}.`);
+    else {
+        out.push("| Agent | Base | Head | Change |", "|---|---:|---:|---:|");
+        for (const a of d.agents)
+            out.push(`| ${(0, markdown_1.mdCode)(a.name, true)} | ${cell(a.base)} | ${cell(a.head)}${a.lowerBound ? " (at least)" : ""} | **${signedTok(a.delta)}**${a.deltaPct !== null && a.delta ? ` (${a.deltaPct > 0 ? "+" : ""}${a.deltaPct}%)` : ""} |`);
+        out.push("");
+        for (const a of d.agents.filter((x) => x.changes.length)) {
+            out.push(`<details><summary>${summaryText(a.name)}: ${a.changes.length + a.moreChanges} changed item(s)</summary>`, "", "| Change | Item | Base | Head |", "|---:|---|---:|---:|");
+            for (const c of a.changes)
+                out.push(`| ${signedTok(c.delta)} | ${chg(c)} ${(0, markdown_1.mdCode)(c.label, true)} | ${cell(c.base)} | ${cell(c.head)} |`);
+            if (a.moreChanges)
+                out.push(`| | … ${a.moreChanges} more | | |`);
+            out.push("", "</details>", "");
+        }
+    }
+    if (over.length)
+        out.push("", `**Over the limit:** ${over.map((a) => `${(0, markdown_1.mdCode)(a.name)} grew ${signedTok(a.delta)} (limit +${(0, fmt_1.fmtTok)(opts.threshold)})`).join(", ")}.`);
+    for (const n of d.notes)
+        out.push("", `_Note: ${(0, fmt_1.sanitize)(n, 200).replace(/[_*`<>[\]]/g, " ")}_`);
+    out.push("", `<sub>Estimates (o200k-based, per-agent factors), not exact for any model. Only this repository's files are scanned. [prehello](https://${card_1.REPO_URL}) ${(0, fmt_1.sanitize)(d.head.tool.version, 20)}</sub>`);
+    return out.join("\n") + "\n";
+}
+function diffJson(d, threshold) {
+    return {
+        tool: { name: "prehello", version: d.head.tool.version },
+        kind: "diff",
+        agents: d.agents,
+        notes: d.notes,
+        ...(threshold !== undefined ? { threshold, overThreshold: overThreshold(d, threshold).map((a) => a.id) } : {}),
+    };
+}
+/** shields.io endpoint JSON (https://shields.io/badges/endpoint-badge). */
+function badgeJson(name, est, lowerBound) {
+    const color = est === null ? "lightgrey" : est < 5000 ? "brightgreen" : est < 15000 ? "green" : est < 30000 ? "yellow" : est < 60000 ? "orange" : "red";
+    return { schemaVersion: 1, label: "context before hello", message: est === null ? "unmeasured" : `${lowerBound ? "≥" : "~"}${(0, fmt_1.fmtTok)(est)} tokens (${name})`, color };
+}
+/** Plain text for inside <summary>: GitHub does not render Markdown there, so no backticks; HTML-escaped. */
+function summaryText(name) {
+    let t = name.replace(/[\u0000-\u001f\u007f-\u009f\u200b-\u200f\u2028-\u202e\u2060-\u2069\ufeff]+/g, " ").replace(/\s+/g, " ").trim();
+    if (t.length > 80)
+        t = t.slice(0, 77) + "...";
+    if (!t)
+        t = "(unnamed)";
+    return t.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;").replace(/"/g, "&quot;").replace(/@/g, "&#64;").replace(/#/g, "&#35;");
+}

package/dist/mcp/configs.js

@@ -31,7 +31,7 @@ function parseMcpMap(map, warn) {
         if (!isObj(v))
             continue;
         const t = typeof v.type === "string" ? v.type : typeof v.transport === "string" ? v.transport : undefined;
-        const transport = t === "http" || t === "streamable-http" ? "http" : t === "sse" ? "sse" : t === "stdio" || typeof v.command === "string" ? "stdio" : typeof v.url === "string" ? "http" : "unknown";
+        const transport = t === "http" || t === "streamable-http" ? "http" : t === "sse" ? "sse" : t === "stdio" || t === "local" || typeof v.command === "string" ? "stdio" : typeof v.url === "string" || typeof v.httpUrl === "string" ? "http" : "unknown";
         out.push({ name: (0, fmt_1.sanitize)(name, 60), transport, pkg: guessPkg(v), disabled: v.disabled === true || v.enabled === false });
     }
     return out;

package/dist/model.js

@@ -2,10 +2,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.LOAD_MODES = exports.ALL_AGENTS = exports.AGENT_NAMES = exports.SCHEMA_VERSION = void 0;
 exports.SCHEMA_VERSION = 1;
-exports.AGENT_NAMES = {
-    "claude-code": "Claude Code",
-    cursor: "Cursor",
-    codex: "Codex",
-};
-exports.ALL_AGENTS = ["claude-code", "cursor", "codex"];
+var agents_1 = require("./agents");
+Object.defineProperty(exports, "AGENT_NAMES", { enumerable: true, get: function () { return agents_1.AGENT_NAMES; } });
+Object.defineProperty(exports, "ALL_AGENTS", { enumerable: true, get: function () { return agents_1.ALL_AGENTS; } });
 exports.LOAD_MODES = ["always", "conditional", "deferred", "on-demand", "unknown"];

package/dist/render/card.js

@@ -1,6 +1,6 @@
 "use strict";
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.CARD_FONT = exports.THEMES = exports.AT_LEAST = exports.MIN_CUT_HEADLINE_TOKENS = void 0;
+exports.CARD_FONT = exports.THEMES = exports.AT_LEAST = exports.MIN_CUT_HEADLINE_TOKENS = exports.REPO_URL = void 0;
 exports.buildHeadline = buildHeadline;
 exports.assertHeadline = assertHeadline;
 exports.pickCardAgent = pickCardAgent;
@@ -10,6 +10,8 @@ const fmt_1 = require("../util/fmt");
 const treemap_1 = require("./treemap");
 const savings_1 = require("../analyze/savings");
 const secrets_1 = require("../util/secrets");
+const agents_1 = require("../agents");
+exports.REPO_URL = "github.com/PIYUSH-KUMAR1809/prehello";
 /** A saving smaller than this (estimated tokens) is noise, not a headline. Same threshold as the cut list itself. */
 exports.MIN_CUT_HEADLINE_TOKENS = 500;
 /** Wording of the lower-bound marker. Words, not a glyph: "≥" is missing from several fallback fonts used by SVG renderers. */
@@ -185,7 +187,7 @@ function renderCard(r, a, o) {
     // footer: wrap on segment boundaries, anchored to the bottom edge
     const unvalidated = a.profile.id.startsWith("claude") || a.profile.id === "cursor-mixed" || a.profile.id === "heuristic";
     const footSize = og ? 17 : 20;
-    const footLines = wrapSegments(["prehello", "local-only scan", `estimate band: ${a.profile.band}${unvalidated ? " (unvalidated)" : ""}`, `window ${(0, fmt_1.fmtWindow)(r.scan.window)}`, ...(r.partial ? ["PARTIAL scan"] : [])], maxCharsFor(innerW, footSize));
+    const footLines = wrapSegments(["npx prehello", exports.REPO_URL, "local-only scan", `estimate band: ${a.profile.band}${unvalidated ? " (unvalidated)" : ""}`, `window ${(0, fmt_1.fmtWindow)(r.scan.window)}`, ...(r.partial ? ["PARTIAL scan"] : [])], maxCharsFor(innerW, footSize));
     const footLineH = footSize + 6;
     const footTop = H - (og ? 30 : 44) - (footLines.length - 1) * footLineH;
     // Layout flows upward from the footer so a wrapped (2-3 line) footer can never collide with the chips or the treemap.
@@ -285,8 +287,7 @@ function renderCard(r, a, o) {
     const unm = a.totals.always.unmeasuredItems + a.totals.deferred.unmeasuredItems;
     if (unm)
         chips.push(`unmeasured: ${unm} item${unm > 1 ? "s" : ""}`);
-    if (a.id === "cursor")
-        chips.push("unmeasured: User Rules");
+    chips.push(...((0, agents_1.agentMeta)(a.id).cardChips ?? []));
     if (r.partial)
         chips.unshift("partial scan");
     let cx = pad;

package/dist/render/markdown.js

@@ -124,7 +124,7 @@ function headlineOf(r, o) {
 function footer(r, shown) {
     const unvalidated = shown.some((a) => a.profile.id.startsWith("claude") || a.profile.id === "cursor-mixed" || a.profile.id === "heuristic");
     const bands = [...new Set(shown.filter((a) => a.totals.always.measuredItems).map((a) => `${a.name} ${ascii(a.profile.band)}`))];
-    return `_Generated by prehello ${r.tool.version} on ${r.generatedAt.slice(0, 10)}. Local-only scan: no paths or file contents are included. All token counts are estimates (${r.scan.tokenizer.method === "bpe-o200k" ? "o200k BPE with per-agent scaling priors" : "chars/3.5 heuristic"}${bands.length ? "; band " + bands.join(", ") : ""}${unvalidated ? ", unvalidated" : ""}), not exact for any model. Window ${(0, fmt_1.fmtWindow)(r.scan.window)}.${r.partial ? " PARTIAL scan: every total is a lower bound." : ""}_`;
+    return `_Generated by prehello ${r.tool.version} on ${r.generatedAt.slice(0, 10)}. Local-only scan: no paths or file contents are included. All token counts are estimates (${r.scan.tokenizer.method === "bpe-o200k" ? "o200k BPE with per-agent scaling factors" : "chars/3.5 heuristic"}${bands.length ? "; band " + bands.join(", ") : ""}${unvalidated ? ", unvalidated" : ""}), not exact for any model. Window ${(0, fmt_1.fmtWindow)(r.scan.window)}.${r.partial ? " PARTIAL scan: every total is a lower bound." : ""} Try it: \`npx prehello\` (https://${card_1.REPO_URL})._`;
 }
 function renderMarkdown(r, o) {
     const shown = r.agents.filter((a) => !a.presentInRepoOnly);

package/dist/render/sarif.js

@@ -337,7 +337,7 @@ function toSarif(r, o) {
                             name: d.name,
                             shortDescription: { text: d.short },
                             fullDescription: { text: d.full },
-                            help: { text: `${d.help}\nSee the SARIF section of the prehello README for the rule list.`, markdown: `${d.help}\n\nSee the SARIF section of the prehello README for the rule list.` },
+                            help: { text: `${d.help}\nThe rule list is in docs/REFERENCE.md (SARIF) in the prehello repository.`, markdown: `${d.help}\n\nThe rule list is in docs/REFERENCE.md (SARIF) in the prehello repository.` },
                             defaultConfiguration: { level: d.level },
                             properties: { tags: ["context", "tokens", "coding-agents"], precision: d.precision },
                         })),
@@ -357,7 +357,7 @@ const round1 = (x) => Math.round(x * 10) / 10;
 function runProps(r, shown, o, noFiles) {
     return {
         prehello: {
-            note: "All token numbers are estimates (o200k BPE with per-agent scaling priors), not exact for any model. An unmeasured item has no number, never 0. A total marked isLowerBound is 'at least'.",
+            note: "All token numbers are estimates (o200k BPE with per-agent scaling factors), not exact for any model. An unmeasured item has no number, never 0. A total marked isLowerBound is 'at least'.",
             tokenizer: r.scan.tokenizer.method,
             claudeProfile: r.scan.claudeProfile,
             windowTokens: r.scan.window,

package/dist/render/terminal.js

@@ -5,6 +5,7 @@ exports.headlineAgents = headlineAgents;
 exports.renderTerminal = renderTerminal;
 exports.renderRubric = renderRubric;
 const fmt_1 = require("../util/fmt");
+const agents_1 = require("../agents");
 const redact_1 = require("../redact");
 const grade_1 = require("../analyze/grade");
 const MODE_TITLES = {
@@ -14,19 +15,6 @@ const MODE_TITLES = {
     "on-demand": "ON-DEMAND (only if invoked)",
     unknown: "UNKNOWN LOAD BEHAVIOUR",
 };
-const LOOKED_FOR = [
-    ["Claude Code", "~/.claude, ~/.claude.json, `claude` on PATH; CLAUDE.md, .claude/, .mcp.json in the project"],
-    ["Cursor", "~/.cursor, Cursor.app, `cursor` on PATH; .cursor/, .cursorrules, AGENTS.md in the project"],
-    ["Codex", "~/.codex, `codex` on PATH; AGENTS.md, .codex/, .agents/skills in the project"],
-];
-const WHERE_LOOKED = {
-    "claude-code": "CLAUDE.md (project, parent folders, ~/.claude), .claude/{rules,skills,commands}, plugins, .mcp.json, ~/.claude.json",
-    cursor: ".cursor/rules (incl. nested), AGENTS.md, .cursorrules, skills folders, ~/.cursor/{skills,plugins,mcp.json}, .cursor/mcp.json",
-    codex: "AGENTS.md chain, ~/.codex/{AGENTS.md,config.toml}, .agents/skills, ~/.agents/skills, .codex/config.toml",
-};
-const AGENT_HINT = {
-    cursor: "Cursor's User Rules live inside the app, not on disk. Paste them into a file and run with --paste-user-rules <file> to count them.",
-};
 const KIND_NOUN = {
     instructions: ["instruction file", "instruction files"],
     rule: ["rule", "rules"],
@@ -170,14 +158,14 @@ function renderTerminal(r, o) {
     const win = (0, fmt_1.fmtWindow)(r.scan.window);
     const shown = headlineAgents(r);
     out.push(`${p.bold("prehello " + r.tool.version)} · ${dp(r.scan.cwd) === "." ? "this folder" : (0, fmt_1.shortenLeft)((0, fmt_1.sanitize)(dp(r.scan.cwd), 400), 60)} + user config · window ${win}`);
-    out.push(...(0, fmt_1.wrap)(`Token counts are ESTIMATES (${r.scan.tokenizer.method === "bpe-o200k" ? "o200k BPE with per-agent scaling priors" : "chars/3.5 heuristic"}), shown with ~. Not exact for any model.`, W).map((l) => p.dim(l)));
+    out.push(...(0, fmt_1.wrap)(`Token counts are ESTIMATES (${r.scan.tokenizer.method === "bpe-o200k" ? "o200k BPE with per-agent scaling factors" : "chars/3.5 heuristic"}), shown with ~. Not exact for any model.`, W).map((l) => p.dim(l)));
     out.push("");
     if (r.agents.length === 0) {
         out.push(`No coding agents detected on this machine, and no agent files found in ${dp(r.scan.cwd) === "." ? "this folder" : dp(r.scan.cwd)}.`);
         out.push("");
         out.push(p.bold("prehello looked for"));
-        for (const [n, t] of LOOKED_FOR)
-            out.push(...(0, fmt_1.wrap)(t, W, `  ${n.padEnd(12)} `, " ".repeat(15)));
+        for (const m of agents_1.AGENTS)
+            out.push(...(0, fmt_1.wrap)(m.lookedFor, W, `  ${m.name.padEnd(15)} `, " ".repeat(18)));
         out.push("");
         out.push(p.bold("Try"));
         out.push("  prehello ~/code/my-project       scan a project folder (default: the current directory)");
@@ -269,9 +257,10 @@ function renderTerminal(r, o) {
         if (!a.items.length) {
             const by = a.detectedBy.length ? ` (via ${a.detectedBy.join(", ")})` : "";
             out.push(p.bold(`   ${a.name} is installed${by}, but nothing that loads before hello was found.`));
-            out.push(...(0, fmt_1.wrap)(`This is NOT a score of 0 and NOT proof the agent loads nothing: it still loads its own system prompt and tools. Looked in: ${WHERE_LOOKED[a.id]}.`, W, "   ", "   ").map((l) => p.dim(l)));
-            if (AGENT_HINT[a.id])
-                out.push(...(0, fmt_1.wrap)(AGENT_HINT[a.id], W, "   ", "   ").map((l) => p.dim(l)));
+            out.push(...(0, fmt_1.wrap)(`This is NOT a score of 0 and NOT proof the agent loads nothing: it still loads its own system prompt and tools. Looked in: ${(0, agents_1.agentMeta)(a.id).whereLooked}.`, W, "   ", "   ").map((l) => p.dim(l)));
+            const hint = (0, agents_1.agentMeta)(a.id).hint;
+            if (hint)
+                out.push(...(0, fmt_1.wrap)(hint, W, "   ", "   ").map((l) => p.dim(l)));
         }
         if (a.mcpServers.length) {
             const names = a.mcpServers.map((m) => m.name);

package/dist/safe-write.js

@@ -26,6 +26,8 @@ function looksLikePrehelloOutput(headRaw, kind) {
     const head = kind === "png" ? headRaw : headRaw.replace(/^(?:\ufeff|\u00ef\u00bb\u00bf)/, "");
     if (kind === "md")
         return head.startsWith(exports.MD_MARKER_PREFIX);
+    if (kind === "badge")
+        return /^\s*\{\s*"schemaVersion":\s*1,\s*"label":\s*"context before hello"/.test(head);
     // `head` is decoded as latin1, so a PNG's bytes survive: signature, then the tEXt "Software: prehello" chunk tagPng() adds right after IHDR.
     if (kind === "png")
         return head.startsWith("\u0089PNG\r\n\u001a\n") && /tEXtSoftware\u0000prehello /.test(head);

package/dist/scan.js

@@ -7,20 +7,14 @@ const context_1 = require("./context");
 const fmt_1 = require("./util/fmt");
 const counter_1 = require("./tokens/counter");
 const model_1 = require("./model");
-const claude_code_1 = require("./adapters/claude-code");
-const cursor_1 = require("./adapters/cursor");
-const codex_1 = require("./adapters/codex");
+const adapters_1 = require("./adapters");
 const totals_1 = require("./analyze/totals");
 const findings_1 = require("./analyze/findings");
 const grade_1 = require("./analyze/grade");
 const savings_1 = require("./analyze/savings");
 const summary_1 = require("./analyze/summary");
-exports.VERSION = "0.1.0";
-const ADAPTERS = {
-    "claude-code": { detect: claude_code_1.detectClaude, discover: claude_code_1.discoverClaude },
-    cursor: { detect: cursor_1.detectCursor, discover: cursor_1.discoverCursor },
-    codex: { detect: codex_1.detectCodex, discover: codex_1.discoverCodex },
-};
+const version_1 = require("./version");
+Object.defineProperty(exports, "VERSION", { enumerable: true, get: function () { return version_1.VERSION; } });
 exports.NOT_MEASURABLE = [
     "the agent's built-in system prompt and built-in tool definitions",
     "server-side / org rules, Cursor User & Team Rules, claude.ai connectors",
@@ -38,7 +32,7 @@ function scan(input) {
     for (const id of model_1.ALL_AGENTS) {
         if (forced && !forced.includes(id))
             continue;
-        const ad = ADAPTERS[id];
+        const ad = adapters_1.ADAPTERS[id];
         const det = ad.detect(ctx);
         const inUse = !!forced || det.detectedBy.length > 0;
         const repoOnly = !inUse && det.repoFiles.length > 0;
@@ -91,7 +85,7 @@ function scan(input) {
     const when = epoch && /^\d+$/.test(epoch) ? new Date(parseInt(epoch, 10) * 1000) : opts.now;
     return {
         schemaVersion: model_1.SCHEMA_VERSION,
-        tool: { name: "prehello", version: exports.VERSION },
+        tool: { name: "prehello", version: version_1.VERSION },
         generatedAt: when.toISOString(),
         scan: { cwd: opts.cwd, window: opts.window, tokenizer: { method: counter.method, note: counter.note }, claudeProfile: opts.claudeProfile, ...(ctx.fixturesSkipped > 0 ? { fixturesSkipped: ctx.fixturesSkipped } : {}) },
         agents,

package/dist/text/imports.js

@@ -51,12 +51,17 @@ function expandImports(fileText, filePath, home, allowed, opts) {
                     skipped.push({ spec: "(more)", reason: `more than ${MAX_CANDIDATES} @imports; the rest were not followed` });
                 continue;
             }
+            if (opts.relativeOnly && (spec.startsWith("~") || node_path_1.default.isAbsolute(spec) || /^[A-Za-z]:[\\/]/.test(spec))) {
+                skipped.push({ spec, reason: "absolute or ~/ path (not loaded by this agent)" });
+                continue;
+            }
             const target = spec.startsWith("~/") ? node_path_1.default.join(home, spec.slice(2)) : node_path_1.default.resolve(node_path_1.default.dirname(from), spec);
             const real = (0, fs_safe_1.realpathOrSelf)(target);
             if (seen.has(real))
                 continue;
-            if (depth >= MAX_DEPTH) {
-                skipped.push({ spec, reason: "depth>4" });
+            const maxDepth = opts.maxDepth ?? MAX_DEPTH;
+            if (depth >= maxDepth) {
+                skipped.push({ spec, reason: `depth>${maxDepth}` });
                 continue;
             }
             if (!allowed(real)) {

package/dist/tokens/families.js

@@ -1,13 +1,30 @@
 "use strict";
+var __importDefault = (this && this.__importDefault) || function (mod) {
+    return (mod && mod.__esModule) ? mod : { "default": mod };
+};
 Object.defineProperty(exports, "__esModule", { value: true });
+exports.claudeCalibration = claudeCalibration;
 exports.profileFor = profileFor;
+const node_fs_1 = __importDefault(require("node:fs"));
+const node_path_1 = __importDefault(require("node:path"));
+const agents_1 = require("../agents");
+let calCache;
+/** Claude ratios from spec/calibration.json (published count_tokens measurements; see docs/ACCURACY.md). */
+function claudeCalibration() {
+    if (!calCache) {
+        const j = JSON.parse(node_fs_1.default.readFileSync(node_path_1.default.join(__dirname, "..", "..", "spec", "calibration.json"), "utf8"));
+        calCache = j.claude;
+    }
+    return calCache;
+}
 function sym(raw, factor, bandPct) {
     const est = Math.round(raw * factor);
     return { est, low: Math.round(est * (1 - bandPct)), high: Math.round(est * (1 + bandPct)) };
 }
+const pct = (x) => `±${Math.round(x * 100)}%`;
 /**
- * Priors from PRODUCT_PLAN §7.2. All Claude/heuristic factors and bands are UNVALIDATED priors until
- * calibrated against ground truth (see docs/accuracy.md). The OpenAI band is for the BPE itself.
+ * Claude factors come from spec/calibration.json (published ratios, still unvalidated: n is small). Codex is o200k itself.
+ * "mixed" agents (Cursor, Copilot) let the user pick the model, so they span GPT-like to Claude-like. Gemini has no measured ratio.
  */
 function profileFor(agent, method, claude) {
     if (method === "heuristic") {
@@ -19,28 +36,43 @@ function profileFor(agent, method, claude) {
             unvalidated: true,
         };
     }
-    if (agent === "codex") {
+    const family = (0, agents_1.agentMeta)(agent).tokenFamily;
+    if (family === "openai") {
         return { id: "openai-o200k", label: "o200k BPE (OpenAI-family)", band: "±3%", apply: (raw) => sym(raw, 1.0, 0.03), unvalidated: false };
     }
-    const f = claude === "current" ? 1.5 : 1.2;
+    const cal = claudeCalibration();
+    const scale = claude === "current" ? 1 : cal.legacyRatio.value;
+    const f = (k) => Math.round(cal.byKind[k].factor * scale * 1000) / 1000;
+    const b = (k) => cal.byKind[k].band;
     const claudeProfile = {
         id: `claude-${claude}`,
-        label: `o200k × ${f} (Claude-${claude} prior)`,
-        band: "±20%",
-        apply: (raw) => sym(raw, f, 0.2),
-        unvalidated: true,
+        label: `o200k × ${f("markdown")} (text) / × ${f("json")} (JSON), Claude-${claude} ratios from published count_tokens measurements`,
+        band: pct(b("markdown")),
+        apply: (raw, kind = "markdown") => sym(raw, f(kind), b(kind)),
+        unvalidated: cal.unvalidated,
     };
-    if (agent === "claude-code")
+    if (family === "claude")
         return claudeProfile;
-    // Cursor: model not discoverable from disk -> span GPT-like (×1.0, −3%) to Claude-like (×f, +20%); est = midpoint.
+    if (family === "gemini") {
+        // No published Gemini-vs-o200k ratio was found (docs/ACCURACY.md), so prehello does not invent one: the estimate is the
+        // raw o200k count and the band is the same GPT-like-to-Claude-like spread used for "model unknown". NOT a Gemini measurement.
+        return {
+            id: "gemini-unvalidated",
+            label: `o200k × 1.0; no measured Gemini ratio, band borrowed from the GPT-like (×1.0) to Claude-like (×${f("markdown")}) spread`,
+            band: `wide, unvalidated: −3% to +${Math.round((f("markdown") * (1 + b("markdown")) - 1) * 100)}%`,
+            apply: (raw, kind = "markdown") => ({ est: Math.round(raw), low: Math.round(raw * 0.97), high: Math.round(raw * f(kind) * (1 + b(kind))) }),
+            unvalidated: true,
+        };
+    }
+    // mixed (Cursor, GitHub Copilot: the model is user-selectable and not discoverable from disk): span GPT-like (×1.0, −3%) to Claude-like (×f, +band); est = midpoint.
     return {
-        id: "cursor-mixed",
-        label: `o200k; model unknown, so midpoint of GPT-like (×1.0) and Claude-like (×${f}) priors`,
-        band: "wide: GPT-like −3% to Claude-like +20%",
-        apply: (raw) => {
+        id: agent === "cursor" ? "cursor-mixed" : "mixed",
+        label: `o200k; model unknown, so midpoint of GPT-like (×1.0) and Claude-like (×${f("markdown")}) ratios`,
+        band: `wide: GPT-like −3% to Claude-like +${Math.round(b("markdown") * 100)}%`,
+        apply: (raw, kind = "markdown") => {
             const lo = Math.round(raw * 0.97);
-            const hi = Math.round(raw * f * 1.2);
-            return { est: Math.round((raw * (1 + f)) / 2), low: lo, high: hi };
+            const hi = Math.round(raw * f(kind) * (1 + b(kind)));
+            return { est: Math.round((raw * (1 + f(kind))) / 2), low: lo, high: hi };
         },
         unvalidated: true,
     };

package/dist/version.js

@@ -0,0 +1,5 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.VERSION = void 0;
+// Generated by scripts/gen-version.mjs from package.json. Do not edit.
+exports.VERSION = "1.0.0";

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "prehello",
-  "version": "0.1.0",
-  "description": "See what your coding agent (Claude Code, Cursor, Codex) loads before you say hello, and what you could cut. Local-only, read-only, no API key.",
+  "version": "1.0.0",
+  "description": "See what your coding agent (Claude Code, Cursor, Codex, GitHub Copilot, Gemini CLI) loads before you say hello, and what you could cut. Local-only, read-only, no API key.",
   "license": "MIT",
   "author": "Piyush Kumar",
   "homepage": "https://github.com/PIYUSH-KUMAR1809/prehello#readme",
@@ -25,11 +25,13 @@
     "THIRD_PARTY_NOTICES.md"
   ],
   "engines": {
-    "node": ">=18"
+    "node": ">=22"
   },
   "scripts": {
     "clean": "node -e \"require('fs').rmSync('dist',{recursive:true,force:true})\"",
+    "prebuild": "node scripts/gen-version.mjs",
     "build": "tsc -p tsconfig.build.json",
+    "pretypecheck": "node scripts/gen-version.mjs",
     "typecheck": "tsc -p tsconfig.json --noEmit",
     "test": "npm run build && vitest run",
     "prepack": "npm run clean && npm run build",
@@ -39,6 +41,8 @@
     "claude-code",
     "cursor",
     "codex",
+    "github-copilot",
+    "gemini-cli",
     "agents-md",
     "ai-agents",
     "context-window",
@@ -55,7 +59,7 @@
     "smol-toml": "^1.9.1"
   },
   "devDependencies": {
-    "@types/node": "^20.19.43",
+    "@types/node": "^22.19.0",
     "@types/picomatch": "^4.0.3",
     "ajv": "^8.20.0",
     "ajv-formats": "^3.0.1",

package/spec/calibration.json

@@ -0,0 +1,33 @@
+{
+  "version": 1,
+  "note": "Claude-to-o200k token ratios from PUBLISHED Anthropic count_tokens measurements (not measured by prehello). factor = median of the published points for that content kind; band = relative half-width used for low/high. Unvalidated until n >= 100 own measurements (scripts/calibrate-anthropic.mjs). See docs/ACCURACY.md.",
+  "claude": {
+    "measuredAt": "2026-10-23",
+    "tokenizer": "Claude 4.7 and later (Opus/Sonnet/Haiku 5.5 give identical counts per both sources)",
+    "n": 10,
+    "unvalidated": true,
+    "byKind": {
+      "markdown": {
+        "factor": 1.43,
+        "band": 0.15,
+        "points": [
+          { "ratio": 1.486, "what": "English prose reference text, 430 o200k tokens", "source": "https://measuredlab.com/tools/token-counter/" },
+          { "ratio": 1.38, "what": "5 Markdown/reST files, 44,621 o200k tokens", "source": "https://aitoollabs.dev/posts/count-tokens-in-codebase/" }
+        ]
+      },
+      "json": {
+        "factor": 1.35,
+        "band": 0.15,
+        "points": [
+          { "ratio": 1.282, "what": "JSON reference text, 535 o200k tokens", "source": "https://measuredlab.com/tools/token-counter/" },
+          { "ratio": 1.42, "what": "5 YAML/JSON/TOML files, 3,391 o200k tokens", "source": "https://aitoollabs.dev/posts/count-tokens-in-codebase/" }
+        ]
+      }
+    },
+    "legacyRatio": {
+      "value": 0.82,
+      "what": "Sonnet 4.6 (pre-4.7 tokenizer) used 18% fewer tokens than 4.7+ on the same 33 files",
+      "source": "https://aitoollabs.dev/posts/count-tokens-in-codebase/"
+    }
+  }
+}

package/spec/cursor.json

@@ -1,7 +1,7 @@
 {
   "agent": "cursor",
   "verifiedAt": "2026-10-25",
-  "verifiedNote": "Rule modes, skill locations (native + compat dirs), skill frontmatter (paths/globs, disable-model-invocation), plugin manifest/default component dirs, and lazy MCP loading (tool names static, descriptions on demand) are from Cursor's docs and blog (fetched 2026-10-25). UNVERIFIED (marked assumed): how Cursor injects agent-requested rule descriptions and the skills listing, ~/.cursor/skills-cursor (observed on a real Mac only), the plugin cache/marketplace on-disk layout and install state, and whether nested .cursor/rules apply (docs and forum disagree).",
+  "verifiedNote": "Rule modes, skill locations (native + compat dirs), skill frontmatter (paths/globs, disable-model-invocation), plugin manifest/default component dirs, and lazy MCP loading (tool names static, descriptions on demand) are from Cursor's docs and blog (fetched 2026-10-25). UNVERIFIED (marked assumed): how Cursor injects agent-requested rule descriptions, ~/.cursor/skills-cursor (observed on a real Mac only), the plugin cache/marketplace on-disk layout and install state, and whether nested .cursor/rules apply (docs and forum disagree). Skills listing (observed 2026-11-03 from forum.cursor.com/t/163761 and /t/163768, Cursor staff replies of 2026-06-22/25): <agent_skills><available_skills><agent_skill fullPath=\"...\">description</agent_skill>, descriptions shortened (80 chars + ellipsis seen in cloud agents) and, past an undocumented cap, path only; modelled as 80-char entries with path-only/full-description bounds.",
   "docs": {
     "rules": "https://cursor.com/docs/rules",
     "mcp": "https://cursor.com/docs/mcp",

package/spec/gemini-cli.json

@@ -0,0 +1,17 @@
+{
+  "agent": "gemini-cli",
+  "verifiedAt": "2026-11-03",
+  "verifiedNote": "GEMINI.md hierarchy (~/.gemini/GEMINI.md, workspace dirs and parents up to context.memoryBoundaryMarkers, default [\".git\"]), just-in-time subdirectory files, context.fileName (string or list), @file imports (max depth 5, code regions ignored), extensions in ~/.gemini/extensions/<name>/gemini-extension.json (contextFileName, mcpServers; settings win on a clash), mcpServers in ~/.gemini/settings.json and .gemini/settings.json (workspace overrides user) and skills (name + description up front) are from the google-gemini/gemini-cli docs (main, fetched 2026-11-03). The docs disagree on subdirectory files (gemini-md.md: just-in-time; reference/configuration.md: eager scan up to 200 dirs); prehello follows gemini-md.md and calls them conditional. ASSUMED: MCP tool schemas loaded up front, extensions enabled, legacy top-level contextFileName. Token accuracy was never validated and no Gemini tokenizer ratio is applied.",
+  "docs": {
+    "geminiMd": "https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/gemini-md.md",
+    "imports": "https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/memport.md",
+    "settings": "https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/configuration.md",
+    "mcp": "https://github.com/google-gemini/gemini-cli/blob/main/docs/reference/configuration.md#mcpservers",
+    "extensions": "https://github.com/google-gemini/gemini-cli/blob/main/docs/extensions/reference.md",
+    "skills": "https://github.com/google-gemini/gemini-cli/blob/main/docs/cli/skills.md"
+  },
+  "caps": {
+    "importMaxDepth": 5,
+    "discoveryMaxDirs": 200
+  }
+}