Fold canonical equivalents into the three identity keys that bypassed the name rule

changelog.d/2692-identity-sites.fixed.md

@@ -0,0 +1 @@
+A sibling Bot name, a member's role label and a memory topic written in either Unicode canonical form are now one identity each: `create_child_bot` refuses a child whose name only spells the sibling's name the other way, a queued task card reaches the member whose role label is the same role, and one topic is one file — a file written under an older spelling is still the topic's file, so nothing already saved is lost ([#2692](https://github.com/ThinkFlowLab/FlyBot/issues/2692)).

core/client.js

@@ -8729,6 +8729,9 @@ body[data-ds-dark-theme] .bt-mw-ask{color:#f0b84a}
 @media (prefers-reduced-motion:reduce){.bt-mw-row{animation:none}}
 `;
 
+// src/shared/identity.js
+var nameKey = (value) => stripInvisible(String(value ?? "").normalize("NFC")).trim().toLowerCase().normalize("NFC");
+
 // src/shared/task-board.js
 var COLUMNS = ["not-started", "in-progress", "done"];
 var PRIORITIES = ["urgent", "high", "normal", "low"];
@@ -8767,8 +8770,8 @@ function compareForWork(a, b) {
   return (b?.updatedAt ?? 0) - (a?.updatedAt ?? 0);
 }
 function sameRole(a, b) {
-  const left = String(a ?? "").trim().toLowerCase();
-  const right = String(b ?? "").trim().toLowerCase();
+  const left = nameKey(a);
+  const right = nameKey(b);
   return left !== "" && left === right;
 }
 function compareForQueue(a, b) {

core/src/host/memory.js

@@ -29,6 +29,7 @@ import { currentParticipants } from './room-membership.js'
 import { defineTool } from './tool-schema.js'
 import { clip, codePoints, splitRef, unquote } from './text.js'
 import { cut, fits } from '../shared/text-boundary.js'
+import { nameKey } from '../shared/identity.js'
 import { output } from './tools.js'
 import { retiredBot } from './store.js'
 import { botNode, buildGraph, buildIndex, relatedOf, searchIndex, topicNode } from './memory-search.js'
@@ -106,9 +107,11 @@ export function cleanEntry(raw) {
   return splitMeta(line).text
 }
 
-// A flat file name: case folds on some file systems, so names are lower case.
+// A flat file name: case folds on some file systems, so names are lower case. Since #2692 the name
+// is also the shared name key, so canonical equivalents are one topic and one file, and a
+// decomposed Latin topic (refused while U+0301 was not a letter) is accepted as its NFC form.
 export function cleanTopic(raw) {
-  const name = String(raw ?? '').trim().replace(/^\[\[|\]\]$/g, '').replace(/\.md$/i, '').trim().toLowerCase().replace(/\s+/g, '-')
+  const name = nameKey(raw).replace(/^\[\[|\]\]$/g, '').replace(/\.md$/i, '').replace(/\s+/g, '-')
   return /^[\p{L}\p{N}][\p{L}\p{N}_-]{0,47}$/u.test(name) && name !== 'memory' ? name : undefined
 }
 
@@ -511,23 +514,31 @@ export function install(rt) {
       return ''
     }
   }
-  const topicNamesIn = (dir) => {
+  // A topic file's name is the topic, but not always the key: a file written before #2692 keeps the
+  // spelling it was written with (a decomposed Hangul syllable among them), so the topic is the key
+  // computed from that name and the name itself is still the file recall reads and remember
+  // appends to. When both spellings exist, the one already in key form wins.
+  const topicFilesIn = (dir) => {
     try {
-      return readdirSync(dir).filter(name => name.endsWith('.md') && name !== MAIN).map(name => name.slice(0, -3)).sort()
+      const byTopic = new Map()
+      for (const file of readdirSync(dir).filter(name => name.endsWith('.md') && name !== MAIN).sort()) {
+        const topic = nameKey(file.slice(0, -3))
+        if (!byTopic.has(topic) || file === `${topic}.md`) byTopic.set(topic, file)
+      }
+      return [...byTopic].map(([topic, file]) => ({ topic, file }))
     } catch {
       return []
     }
   }
   const topicNames = (scope) => {
     const dir = scopeDir(scope)
-    return dir === undefined ? [] : topicNamesIn(dir)
+    return dir === undefined ? [] : topicFilesIn(dir).map(file => file.topic)
   }
   // MEMORY.md first, then the topics by name.
   const loadScope = (scope) => {
     const dir = dirOf(scope)
     const open = scopeDir(scope) !== undefined
-    return [undefined, ...(open ? topicNamesIn(dir) : [])].map((topic) => {
-      const file = topic === undefined ? MAIN : `${topic}.md`
+    return [{ topic: undefined, file: MAIN }, ...(open ? topicFilesIn(dir) : [])].map(({ topic, file }) => {
       const content = open ? readText(join(dir, file)) : ''
       return { topic, file, path: join(dir, file), content, ...parseMemory(content) }
     })
@@ -690,7 +701,7 @@ export function install(rt) {
       const next = joinLines(insertEntry(base, line))
       checkSize(target, next, label)
       let index
-      if (topic !== undefined && !main.links.includes(topic)) {
+      if (topic !== undefined && !main.links.some(link => nameKey(link) === topic)) {
         index = joinLines(addLink(main.lines.length > 0 ? main.lines : parseMemory(EMPTY_MAIN).lines, topic))
         checkSize(main, index, label)
       }
@@ -745,7 +756,10 @@ export function install(rt) {
       if (empty) {
         await unlink(old.file.path).catch(error => { if (error?.code !== 'ENOENT') throw error })
         const main = files[0]
-        if (main.links.includes(old.file.topic)) await writeText(main.path, joinLines(removeLink(main.lines, old.file.topic)))
+        // The link was written with whatever spelling the topic had then; remove that one, not a
+        // second entry the key would otherwise leave behind (#2692).
+        const link = main.links.find(one => nameKey(one) === old.file.topic)
+        if (link !== undefined) await writeText(main.path, joinLines(removeLink(main.lines, link)))
       } else {
         await writeText(old.file.path, joinLines(lines))
       }

core/src/host/text.js

@@ -2,6 +2,7 @@
 // cutting and estimating text, and the clock format of chat anchors.
 import { COLORS, EXTRA_COLORS, LEGACY_COLORS, OWN_CHAT_LINE_MAX } from './constants.js'
 import { codePointEnd, codePointStart, cut, fits } from '../shared/text-boundary.js'
+import { nameKey } from '../shared/identity.js'
 import { stripInvisible } from '../shared/mentions.js'
 
 export { cut, stripInvisible }
@@ -21,10 +22,9 @@ export const textOf = content => (Array.isArray(content) ? content : [])
   .join('\n')
   .trim()
 
-// The identity key for a name (#2691): both canonical spellings (NFC and NFD) of one
-// visible name are one key, and so are the invisible characters `trim()` keeps. The NFC
-// pass runs again after lowercasing because a lower can compose a new sequence (U+0130).
-export const normalize = value => stripInvisible(String(value ?? '').normalize('NFC')).trim().toLowerCase().normalize('NFC')
+// The identity key for a name (#2691, #2692): one function in `shared/identity.js`, so the tree's
+// sibling names, a role label and a memory topic compare by the same rule as a Bot name does.
+export const normalize = nameKey
 
 export const unquote = text => String(text ?? '').trim().replace(/^["'“”「『《]+|["'“”」』》]+$/g, '').trim()
 

core/src/shared/identity.js

@@ -0,0 +1,7 @@
+// The identity key for a visible name (#2691, #2692), in one place so the host, a shared module and
+// a name that is also a file name cannot drift apart. Both canonical spellings (NFC and NFD) of one
+// visible name are one key, and so are the invisible format characters `trim()` keeps. The NFC pass
+// runs again after lowercasing because a lower can compose a new sequence (U+0130).
+import { stripInvisible } from './mentions.js'
+
+export const nameKey = value => stripInvisible(String(value ?? '').normalize('NFC')).trim().toLowerCase().normalize('NFC')

core/src/shared/task-board.js

@@ -1,6 +1,7 @@
 // Board columns for the task records in src/host/tasks.js (#1058).
 // The state machine stays the record. A column is how the board reads it.
 // result-ready still waits for acceptance, so it stays In progress.
+import { nameKey } from './identity.js'
 
 export const COLUMNS = ['not-started', 'in-progress', 'done']
 export const PRIORITIES = ['urgent', 'high', 'normal', 'low']
@@ -54,9 +55,11 @@ export function compareForWork(a, b) {
 
 // A role label, matched the way a person reads it (#1247): free text, trimmed, case-insensitive.
 // The label is not a membership role (`admin` / `member`), so nothing here reads that field.
+// Since #2692 the comparison is the shared name key, so either canonical spelling of one visible
+// label is one role and a queued card reaches the carrier that spells it the other way.
 export function sameRole(a, b) {
-  const left = String(a ?? '').trim().toLowerCase()
-  const right = String(b ?? '').trim().toLowerCase()
+  const left = nameKey(a)
+  const right = nameKey(b)
   return left !== '' && left === right
 }
 

core/test/bot-tasks.test.js

@@ -9,7 +9,7 @@ import { tmpdir } from 'node:os'
 import { join } from 'node:path'
 import { test } from 'node:test'
 import { apply } from '../index.js'
-import { columnOf, occupiesSlot } from '../src/shared/task-board.js'
+import { columnOf, occupiesSlot, sameRole } from '../src/shared/task-board.js'
 import { fakeContext } from './fake-ctx.js'
 
 const team = () => ({
@@ -681,3 +681,20 @@ test('only a claimed or working card occupies the member\'s WIP slot (#1744)', (
   assert.equal(occupiesSlot({}), false)
   assert.equal(columnOf({ status: 'result-ready' }), 'in-progress', 'the board reading is unchanged')
 })
+
+// #2692: a role label is one identity whichever canonical spelling it is written in, so a queued
+// card reaches the member whose label spells the same role instead of staying in the pool forever.
+test('#2692 a role label is one identity whichever canonical spelling it uses', () => {
+  const nfc = 'Caf\u00e9'
+  const nfd = 'Cafe\u0301'
+  assert.equal(sameRole(nfc, nfd), true)
+  assert.equal(sameRole(nfd, nfc), true)
+  // the folded spelling is the one the tree plugin hands out as a role label
+  assert.equal(sameRole(` ${nfd} `, nfc), true)
+  // the invisible format characters trim() keeps are part of the same identity
+  assert.equal(sameRole(`${nfd}\u200b`, nfc), true)
+  // a different role still is a different role, and an empty label still matches nothing
+  assert.equal(sameRole(nfd, 'Caf\u00e9 writer'), false)
+  assert.equal(sameRole('', ''), false)
+  assert.equal(sameRole(undefined, nfc), false)
+})

core/test/memory.test.js

@@ -227,6 +227,39 @@ test('topics hold details, linked from MEMORY.md, and go when their last entry d
   assert.equal(read('bots', 'b2', 'MEMORY.md'), '# Memory\n\n## Index\n')
 })
 
+// #2692: a topic is one identity and one file name. Canonically equivalent spellings used to be
+// two topics writing two files, so one topic split into two memories. A decomposed Latin topic is
+// refused by the \p{L} rule, so the collision needs a script whose decomposition stays letters:
+// the Hangul syllable 가 (U+AC00) and U+1100 U+1161.
+test('#2692 a topic is one identity whichever canonical spelling it is written in', async (t) => {
+  const { tool, read, stop, home } = start()
+  t.after(() => stop())
+  assert.equal(cleanTopic('\uac00'), cleanTopic('\u1100\u1161'))
+  assert.equal(cleanTopic('\u1100\u1161'), '\uac00', 'the file stem is the canonical form')
+  await tool('remember', { text: 'Flies from Hangzhou', topic: '\uac00' }, 'b2')
+  // the other spelling names the same topic, and reads the same file
+  assert.match(await tool('recall', { topic: '\u1100\u1161' }, 'b2'), /^Topic \[\[\uac00\]\] in your memory \(\uac00\.md: \d+ of 16000 characters\):\n# \uac00\n\n- Flies from Hangzhou/)
+  assert.match(await tool('remember', { text: 'Prefers window seats', topic: '\u1100\u1161' }, 'b2'), /in topic \[\[\uac00\]\] \(\uac00\.md: \d+ of 16000 characters\)/)
+  assert.deepEqual(readdirSync(join(home, 'memory', 'bots', 'b2')).sort(), ['MEMORY.md', '\uac00.md'])
+  assert.equal(read('bots', 'b2', 'MEMORY.md'), '# Memory\n\n## Index\n- [[\uac00]]\n')
+})
+
+// A topic file written before this change keeps the spelling it was written with. The key is
+// computed from the name, so the file is still the one recall reads and remember appends to.
+test('#2692 a topic saved under another spelling before this change is still the file it names', async (t) => {
+  const { tool, read, stop, home } = start()
+  t.after(() => stop())
+  const dir = join(home, 'memory', 'bots', 'b2')
+  mkdirSync(dir, { recursive: true })
+  writeFileSync(join(dir, 'MEMORY.md'), '# Memory\n\n## Index\n- [[\u1100\u1161]]\n')
+  writeFileSync(join(dir, '\u1100\u1161.md'), '# legs\n\n- Saved the old way [added: 2026-01-01]\n')
+  assert.match(await tool('recall', { topic: '\uac00' }, 'b2'), /^Topic \[\[\uac00\]\] in your memory \(\u1100\u1161\.md: \d+ of 16000 characters\):\n# legs/)
+  assert.match(await tool('remember', { text: 'Added after the fix', topic: '\uac00' }, 'b2'), /in topic \[\[\uac00\]\]/)
+  assert.deepEqual(readdirSync(dir).sort(), ['MEMORY.md', '\u1100\u1161.md'], 'no second file for the same topic')
+  assert.match(read('bots', 'b2', '\u1100\u1161.md'), /- Added after the fix \[/)
+  assert.equal(read('bots', 'b2', 'MEMORY.md'), '# Memory\n\n## Index\n- [[\u1100\u1161]]\n', 'the link it already had is not written twice')
+})
+
 test('Main Bots and group admins write the team memory; only Main Bots touch another Bot\'s', async (t) => {
   const { tool, read, stop } = start()
   t.after(() => stop())

docs/subsystems/core-client.md

@@ -7,7 +7,7 @@ owned_paths:
 
 # core-client
 
-The browser side of the patched `ds-bot` core (`core/`): sources in `core/src/client/`, built into the committed bundle `core/client.js`. Split out of [ds-bot-core](ds-bot-core.md) so a host-only change does not have to edit this page. `client.js` is generated: never hand-edit it, run `npm run build` in `core/` and commit both. The bundle also carries the `core/src/shared/` modules the client imports — `mentions.js` (the `@Name` rule, `nameProblem` and `stripInvisible`), `text-boundary.js`, `remote-label.js`, `approval-waiting.js`, `room-share.js` among them — so a change in one of those rebuilds `client.js` too even when no file under `core/src/client/` was edited; the rules themselves are documented on [core-host](core-host.md#key-files).
+The browser side of the patched `ds-bot` core (`core/`): sources in `core/src/client/`, built into the committed bundle `core/client.js`. Split out of [ds-bot-core](ds-bot-core.md) so a host-only change does not have to edit this page. `client.js` is generated: never hand-edit it, run `npm run build` in `core/` and commit both. The bundle also carries the `core/src/shared/` modules the client imports — `mentions.js` (the `@Name` rule, `nameProblem` and `stripInvisible`), `text-boundary.js`, `remote-label.js`, `approval-waiting.js`, `room-share.js`, `identity.js` (the name key `nameKey`, #2692, which `task-board.js` compares a role label with) among them — so a change in one of those rebuilds `client.js` too even when no file under `core/src/client/` was edited; the rules themselves are documented on [core-host](core-host.md#key-files).
 
 ## Owns