Files
gitmost/packages/mcp/test/unit/json-edit-refuse.test.mjs
T
agent_coder 327737b701 feat(ai-chat): give the in-app agent insertFootnote/insertImage/replaceImage (#410)
The Researcher role wrote 40 literal `^[...]` and zero real footnotes: its
incremental write path (insertNode/editPageText) doesn't parse markdown, and
the footnote-capable tool was MCP-only. Promote three tools from inline
MCP-only to the shared registry so the in-app agent gets them too.

- tool-specs.ts: insertFootnote/insertImage/replaceImage added to
  SHARED_TOOL_SPECS (mcpName/schema/description moved VERBATIM from the inline
  registrations — MCP names + behaviour unchanged for external clients).
- index.ts: the 3 inline registerTool calls become registerShared; drop the
  "MCP-only by design" comments.
- ai-chat-tools.service.ts: register the 3 in-app via sharedTool ->
  client.insertFootnote/insertImage/replaceImage (imageUrl->url,
  attachmentId->oldAttachmentId mapping).
- tool-tiers.ts: insertFootnote -> core (else the original asymmetry recurs —
  footnote tool hidden while editPageText is core); images -> deferred.
- research/{en,ru}.yaml FOOTNOTES: `^[...]` parses ONLY on a whole-markdown
  write (create/update/import); for a pinpoint citation to existing text use
  insertFootnote; via editPageText/insertNode it stays literal.
- json-edit.ts guardrail: an edit_page_text `replace` containing a `^[...]`
  token is refused into failed[] with an insert_footnote hint, mirroring the
  existing formatting-marker refusal. (Slightly broader net than that mirror —
  a literal `^[a-z]` regex class in a replace is also refused; accepted
  defense-in-depth, has a no-false-positive test.)

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-07-07 21:18:01 +03:00

184 lines
8.4 KiB
JavaScript

import { test } from "node:test";
import assert from "node:assert/strict";
import { applyTextEdits } from "../../build/lib/json-edit.js";
// Helpers to build small ProseMirror docs.
const textNode = (text, extra = {}) => ({ type: "text", text, ...extra });
const paragraph = (...children) => ({ type: "paragraph", content: children });
const doc = (...children) => ({ type: "doc", content: children });
// ---------------------------------------------------------------------------
// (i) formattingOnly: find and replace differ ONLY by markdown markers
// (find:"~~x~~" / replace:"x"). The text "x" exists, but the edit is a pure
// formatting toggle -> refused into failed[], nothing applied.
// ---------------------------------------------------------------------------
test("formatting-only edit (strip-toggle) is refused, not applied", () => {
const input = doc(paragraph(textNode("x", { marks: [{ type: "strike" }] })));
const snapshot = JSON.parse(JSON.stringify(input));
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "~~x~~", replace: "x" },
]);
assert.equal(results.length, 0, "nothing applied");
assert.equal(failed.length, 1, "one refused edit");
assert.equal(failed[0].find, "~~x~~");
assert.match(failed[0].reason, /cannot add or remove formatting marks/);
assert.match(failed[0].reason, /patch_node/);
// The document is untouched (the strike mark is preserved).
assert.deepEqual(out, snapshot);
});
// ---------------------------------------------------------------------------
// (ii) formattingOnly via add-bold: a plain `find:"x"` whose `replace:"**x**"`
// only adds balanced markers. stripBalancedWrappers(replace) == find, find !=
// replace -> formattingOnly -> refused (it would write a LITERAL `**x**`).
// ---------------------------------------------------------------------------
test("edit that only adds bold markers around plain text is refused", () => {
const input = doc(paragraph(textNode("x")));
const snapshot = JSON.parse(JSON.stringify(input));
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "x", replace: "**x**" },
]);
assert.equal(results.length, 0, "nothing applied");
assert.equal(failed.length, 1, "one refused edit");
assert.match(failed[0].reason, /cannot add or remove formatting marks/);
// No literal ** was written into the document.
assert.deepEqual(out, snapshot);
});
// ---------------------------------------------------------------------------
// (ii-b) More real formatting toggles are still caught by stripBalancedWrappers.
// ---------------------------------------------------------------------------
test("strike-toggle on a price is refused", () => {
const input = doc(paragraph(textNode("$69", { marks: [{ type: "strike" }] })));
const snapshot = JSON.parse(JSON.stringify(input));
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "~~$69~~", replace: "$69" },
]);
assert.equal(results.length, 0, "nothing applied");
assert.equal(failed.length, 1, "one refused edit");
assert.match(failed[0].reason, /cannot add or remove formatting marks/);
assert.deepEqual(out, snapshot);
});
test("nested-wrapper toggle (~~~~**M5Stack**~~~~ -> **M5Stack**) is refused", () => {
const input = doc(
paragraph(textNode("M5Stack", { marks: [{ type: "bold" }] })),
);
const snapshot = JSON.parse(JSON.stringify(input));
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "~~~~**M5Stack**~~~~", replace: "**M5Stack**" },
]);
assert.equal(results.length, 0, "nothing applied");
assert.equal(failed.length, 1, "one refused edit");
assert.match(failed[0].reason, /cannot add or remove formatting marks/);
assert.deepEqual(out, snapshot);
});
// ---------------------------------------------------------------------------
// (ii-c) REGRESSION: ordinary plain-text edits that the OLD lenient detector
// wrongly refused (false positives) now APPLY — they land in `results`, never
// in `failed`. Each `find` exists verbatim in the built doc.
// ---------------------------------------------------------------------------
test("plain-text edits formerly mis-flagged as formatting now apply", () => {
const cases = [
// trailing-space trim: lenient strip trimmed the space -> equal -> refused.
{ find: "tail ", replace: "tail", before: "head tail more" },
// snake_case: `_case_` looked like `_x_` emphasis to the lenient detector.
{ find: "oldname", replace: "snake_case_name", before: "the oldname here" },
// math: `* 3 *` looked like `*x*` emphasis.
{ find: "X", replace: "2 * 3 * 4", before: "value X end" },
// identifier with underscores.
{ find: "A", replace: "my_var_name", before: "set A now" },
];
for (const c of cases) {
const input = doc(paragraph(textNode(c.before)));
const { results, failed } = applyTextEdits(input, [
{ find: c.find, replace: c.replace },
]);
assert.equal(
failed.length,
0,
`"${c.find}" -> "${c.replace}" must NOT be refused (got: ${JSON.stringify(failed)})`,
);
assert.equal(results.length, 1, `"${c.find}" must apply once`);
assert.equal(results[0].find, c.find);
assert.equal(results[0].replacements, 1);
}
});
// ---------------------------------------------------------------------------
// (iii) Legit typo fix: find has markdown but replace differs in LETTERS and
// has no markers. stripped find != stripped replace AND replace has no markers
// -> neither flag trips -> the edit applies.
// ---------------------------------------------------------------------------
test("typo fix wrapped in markdown still applies (not refused)", () => {
// The document renders "M5Stack Atom Eco" with that span bold (misspelled).
const input = doc(
paragraph(textNode("M5Stack Atom Eco", { marks: [{ type: "bold" }] })),
);
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "**M5Stack Atom Eco**", replace: "M5Stack Atom Echo" },
]);
assert.equal(failed.length, 0, "not refused");
assert.equal(results.length, 1, "applied");
assert.equal(results[0].find, "**M5Stack Atom Eco**");
assert.equal(results[0].replacements, 1);
// It matched via the markdown-strip fallback.
assert.equal(results[0].normalized, true);
// The fix is applied AND the bold mark is preserved (text edit, not a
// formatting change).
const node = out.content[0].content.find((n) => n.text === "M5Stack Atom Echo");
assert.ok(node, "the corrected text node exists");
assert.deepEqual(node.marks, [{ type: "bold" }]);
});
// ---------------------------------------------------------------------------
// (iv) #410 footnote token: a `replace` containing `^[...]` is refused into
// failed[] (it would be written as a LITERAL string, never a real footnote).
// Nothing is applied; the reason points at insert_footnote.
// ---------------------------------------------------------------------------
test("replace containing a `^[...]` footnote token is refused, not applied", () => {
const input = doc(paragraph(textNode("The claim stands.")));
const snapshot = JSON.parse(JSON.stringify(input));
const { doc: out, results, failed } = applyTextEdits(input, [
{ find: "The claim stands.", replace: "The claim stands.^[See source, p.42]" },
]);
assert.equal(results.length, 0, "nothing applied");
assert.equal(failed.length, 1, "one refused edit");
assert.equal(failed[0].find, "The claim stands.");
assert.match(failed[0].reason, /insert_footnote/);
// The document is byte-for-byte untouched — no literal `^[` was written.
assert.deepEqual(out, snapshot);
});
test("a plain replace with no footnote token still applies (no false positive)", () => {
const input = doc(paragraph(textNode("a caret ^ and a bracket ] apart")));
const { results, failed } = applyTextEdits(input, [
{ find: "apart", replace: "separate" },
]);
assert.equal(failed.length, 0, "not refused");
assert.equal(results.length, 1, "applied");
});
// ---------------------------------------------------------------------------
// A plain text fix is unaffected by the refuse logic.
// ---------------------------------------------------------------------------
test("plain find/replace is not refused", () => {
const input = doc(paragraph(textNode("teh cat")));
const { results, failed } = applyTextEdits(input, [
{ find: "teh", replace: "the" },
]);
assert.equal(failed.length, 0);
assert.deepEqual(results, [{ find: "teh", replacements: 1 }]);
});