@gorajing/zuun 0.1.1 → 0.1.3

package/README.md

@@ -90,7 +90,7 @@ For the git post-commit hook, manual captures, and shell pipelines:
 
 ```bash
 npm install -g @gorajing/zuun
-zuun --version   # should print 0.1.1
+zuun --version   # should print 0.1.3
 ```
 
 (The npm package is scoped as `@gorajing/zuun` because `zuun` is too close to an existing unscoped package. The CLI binary is still `zuun` — the scope is just for the registry namespace.)
@@ -195,7 +195,7 @@ zuun edit <id>               open in $EDITOR; re-validate on save; DB untouched
 zuun install-git-hook        install post-commit hook in current repo
 zuun capture-commit          (invoked by hook; not for direct use)
 
-zuun doctor                  health check: disk vs DB, schema, ollama, broken refs, log tail
+zuun doctor                  health check: disk vs DB, schema, ollama, broken refs, audit cadence, log tail
 zuun version                 print version
 zuun help                    show commands
 
@@ -230,7 +230,7 @@ Required fields:
 | `created` | ISO 8601 | UTC |
 | `body`  | string     | One self-contained claim. "Local-first beats cloud-first because…" — not a paragraph. |
 | `kind`  | enum       | `decision` · `observation` · `pattern` · `commitment` · `reference` |
-| `source`| enum       | `claude-code` · `cursor` · `git` · `manual` · `import` |
+| `source`| enum       | `claude-code` · `cursor` · `codex` · `git` · `manual` · `import` |
 
 Optional fields: `stance`, `tags[]`, `related[]`, `confidence`, `origin`, `project`.
 
@@ -248,7 +248,7 @@ Environment variables:
 | `OLLAMA_URL` | `http://127.0.0.1:11434` | Ollama server for embeddings |
 | `ZUUN_EMBED_MODEL` | `nomic-embed-text` | Embedding model name |
 | `ZUUN_SEARCH_BLEND` | `fts=0.45,vec=0.45,recency=0.1` | Hybrid search weights |
-| `ZUUN_MCP_SOURCE` | `claude-code` | Tag for entries created via MCP |
+| `ZUUN_MCP_SOURCE` | `claude-code` | `source` tag for entries created via MCP — any `source` enum value (e.g. `codex` when wired into Codex). Unrecognized values fall back to `claude-code`. |
 | `ZUUN_BIN` | *(set by shim)* | Absolute path to `bin/zuun.js`, used by git hook installer |
 | `EDITOR` | `vi` | Editor for `zuun edit` |
 
@@ -302,7 +302,7 @@ Budget: **warm search <100ms on 1k entries**. Current on M-series Mac: ~2ms hybr
 
 ## Project status
 
-**v0.1.1 released.** Published to npm; plugin installable from GitHub via the marketplace flow. 167 tests green. Full plugin surface verified end-to-end on a real Claude Code session (MCP tools, slash command, SessionStart hook, git post-commit hook). Perf within budget by ~50×.
+**v0.1.3 released.** `codex` is now a first-class entry source: wire `zuun mcp` into Codex (OpenAI CLI) via `[mcp_servers.zuun]` with `ZUUN_MCP_SOURCE=codex`, and Codex-captured memories are attributed correctly instead of falling back to `claude-code`. 203 tests green. Published to npm; plugin installable from GitHub via the marketplace flow, with the full plugin surface verified end-to-end on a real Claude Code session (MCP tools, slash command, SessionStart hook, git post-commit hook). Perf within budget by ~50×. See [CHANGELOG.md](CHANGELOG.md) for release history.
 
 This is pre-1.0 software. Schema is versioned (`schema_version: 2`); breaking changes will get a migration path, not a silent reset.
 

package/dist/capture.js

@@ -9,6 +9,7 @@ const embed_provider_1 = require("./lib/embed-provider");
 const embed_1 = require("./lib/embed");
 const entry_1 = require("./lib/entry");
 const tags_1 = require("./lib/tags");
+const lints_1 = require("./lib/lints");
 const dedup_1 = require("./lib/dedup");
 const project_1 = require("./lib/project");
 const log_1 = require("./lib/log");
@@ -58,6 +59,14 @@ async function capture(argv) {
             (0, log_1.appendLog)("capture.dedup", { id: existing });
             return 0;
         }
+        // Lint on RAW tags before normalization: normalizeTags lowercases, but an
+        // entry ID is uppercase-hex, so a polluting ENT token must be detected and
+        // routed to related here or it becomes undetectable (ENT-260514-4BA0).
+        const routing = (0, lints_1.routeEntTokensFromTags)(opts.tags, []);
+        if (routing.routed.length > 0) {
+            process.stderr.write(`capture: moved entry id(s) from tags to related: ${routing.routed.join(", ")} ` +
+                `(tags are for topics; related links entries)\n`);
+        }
         const id = (0, id_1.newEntryId)(body, now);
         const entry = {
             id,
@@ -65,10 +74,15 @@ async function capture(argv) {
             body,
             kind: opts.kind,
             source: "manual",
-            tags: (0, tags_1.normalizeTags)(opts.tags),
-            related: [],
+            tags: (0, tags_1.normalizeTags)(routing.tags),
+            related: routing.related,
             project: (0, project_1.resolveProject)(),
         };
+        // short-decision-no-tags lint; provenance ENT-260514-75A3 (see lints.ts).
+        if ((0, lints_1.isShortDecisionNoTags)(entry)) {
+            process.stderr.write(`capture: ${id} is a short decision with no tags — likely an accidental ` +
+                `auto-capture; 'zuun forget ${id}' if it is not a durable choice\n`);
+        }
         (0, entry_io_1.writeEntry)(entry);
         (0, store_1.upsertEntry)(db, entry);
         const vec = await embed_provider_1.defaultProvider.embed(body);

package/dist/cli.js

@@ -45,7 +45,7 @@ const doctor_1 = require("./lib/doctor");
 const log_1 = require("./lib/log");
 // Kept in sync with package.json on release. If this drifts again, switch to
 // a runtime read of package.json (deferred: avoiding rootDir + tsc complexity).
-const VERSION = "0.1.1";
+const VERSION = "0.1.3";
 const HELP = `usage: zuun <command>
 
 commands:
@@ -60,7 +60,7 @@ commands:
   edit ID      open an entry in $EDITOR and re-validate on save
   install-git-hook  install a git post-commit hook in the current repo
   capture-commit    capture the latest git commit (invoked by the post-commit hook)
-  doctor       health check: entries, db, ollama, broken refs
+  doctor       health check: entries, db, ollama, broken refs, audit cadence
   version      print the zuun version
   help         show this message
 `;

package/dist/lib/audit.js

@@ -0,0 +1,37 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.AUDIT_THRESHOLD = void 0;
+exports.auditStatus = auditStatus;
+/**
+ * Re-audit cadence. ENT-260420-4BDD committed to a corpus re-audit "at ~50
+ * entries"; ENT-260514-98FC observed that passive scheduling let the corpus
+ * run 67 entries past that point before an audit fired. doctor enforces it.
+ */
+exports.AUDIT_THRESHOLD = 50;
+/** Entries tagged this are treated as the marker an audit was performed. */
+const AUDIT_TAG = "audit";
+/**
+ * Audit state derived from the corpus itself — no sidecar file to drift out of
+ * sync. An audit run leaves entries tagged `audit`; the newest one marks when
+ * the last audit happened. Entries created after it are the un-audited backlog.
+ */
+function auditStatus(entries) {
+    let last = null;
+    for (const e of entries) {
+        if (e.tags.includes(AUDIT_TAG) && (last === null || e.created > last.created)) {
+            last = e;
+        }
+    }
+    if (last === null) {
+        const sinceAudit = entries.length;
+        return { lastAuditId: null, lastAuditAt: null, sinceAudit, overdue: sinceAudit >= exports.AUDIT_THRESHOLD };
+    }
+    const cutoff = last.created;
+    const sinceAudit = entries.reduce((n, e) => (e.created > cutoff ? n + 1 : n), 0);
+    return {
+        lastAuditId: last.id,
+        lastAuditAt: last.created,
+        sinceAudit,
+        overdue: sinceAudit >= exports.AUDIT_THRESHOLD,
+    };
+}

package/dist/lib/doctor.js

@@ -6,6 +6,7 @@ const entry_io_1 = require("./entry-io");
 const store_1 = require("./store");
 const embed_provider_1 = require("./embed-provider");
 const log_1 = require("./log");
+const audit_1 = require("./audit");
 async function runDoctor() {
     const db = (0, db_1.openDb)();
     const lines = [];
@@ -29,6 +30,23 @@ async function runDoctor() {
             }
         }
         lines.push(`broken related refs: ${broken}`);
+        // Audit-cadence enforcement; provenance: ENT-260420-4BDD, ENT-260514-98FC
+        // (see audit.ts). Spec IDs stay in comments — output must stand alone for
+        // users who don't have this corpus.
+        const audit = (0, audit_1.auditStatus)(inDb);
+        if (audit.overdue) {
+            healthy = false;
+            const baseline = audit.lastAuditId
+                ? `last audit ${audit.lastAuditId} (${audit.lastAuditAt})`
+                : "no audit on record";
+            lines.push(`WARN: audit overdue — ${audit.sinceAudit} entries since ${audit.lastAuditId ? "last audit" : "first capture"} (threshold ${audit_1.AUDIT_THRESHOLD}); ${baseline}; run a corpus re-audit`);
+        }
+        else if (audit.lastAuditId) {
+            lines.push(`audit: ${audit.sinceAudit} entries since last audit ${audit.lastAuditId} (threshold ${audit_1.AUDIT_THRESHOLD})`);
+        }
+        else {
+            lines.push(`audit: ${audit.sinceAudit} entries, never audited (threshold ${audit_1.AUDIT_THRESHOLD})`);
+        }
         const vec = await embed_provider_1.defaultProvider.embed("doctor-check");
         lines.push(`ollama: ${vec ? "up" : "down"}`);
         if (!vec)

package/dist/lib/entry.js

@@ -28,6 +28,7 @@ exports.EntryKind = zod_1.z.enum([
 exports.EntrySource = zod_1.z.enum([
     "claude-code", // auto-captured from a Claude Code session
     "cursor", // auto-captured from Cursor
+    "codex", // captured from a Codex (OpenAI CLI) session via the MCP server
     "git", // captured from a commit / PR / diff
     "manual", // user typed it in via remember() or CLI
     "import", // bulk-imported from another system

package/dist/lib/lints.js

@@ -0,0 +1,62 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.SHORT_DECISION_MAX = void 0;
+exports.routeEntTokensFromTags = routeEntTokensFromTags;
+exports.isShortDecisionNoTags = isShortDecisionNoTags;
+const entry_1 = require("./entry");
+/**
+ * Capture-path lints derived from the 2026-05-14 corpus audit.
+ *
+ * These run on the *raw* capture input, before tag normalization — by design.
+ * `normalizeTags` lowercases, but ENTRY_ID_REGEX is uppercase-hex only, so an
+ * ENT token routed after normalization would be undetectable (ENT-260514-4BA0).
+ */
+function isEntToken(raw) {
+    return entry_1.ENTRY_ID_REGEX.test(raw.trim().toUpperCase());
+}
+/**
+ * ENT-token-in-tags lint (ENT-260514-4BA0). Tag-index pollution: entry IDs
+ * listed under `tags:` instead of `related:`. Strip ENT-shaped tokens from
+ * tags and route them to related as canonical uppercase IDs.
+ */
+function routeEntTokensFromTags(tags, related) {
+    const keptTags = [];
+    const relatedSeen = new Set(related.map((r) => r.toUpperCase()));
+    const mergedRelated = [...related];
+    const routedSeen = new Set();
+    const routed = [];
+    for (const t of tags) {
+        if (!isEntToken(t)) {
+            keptTags.push(t);
+            continue;
+        }
+        const id = t.trim().toUpperCase();
+        // routed = every ENT token stripped from tags (deduped), even if it was
+        // already in related — the polluting tag was still removed and should be
+        // reported. mergedRelated only gains IDs not already linked.
+        if (!routedSeen.has(id)) {
+            routedSeen.add(id);
+            routed.push(id);
+        }
+        if (!relatedSeen.has(id)) {
+            relatedSeen.add(id);
+            mergedRelated.push(id);
+        }
+    }
+    return { tags: keptTags, related: mergedRelated, routed };
+}
+/**
+ * A decision shorter than this (trimmed) with no tags reads as transient
+ * state, not a durable choice — likely an accidental auto-capture.
+ */
+exports.SHORT_DECISION_MAX = 80;
+/**
+ * short-decision-no-tags lint (ENT-260514-75A3). A `kind=decision` entry whose
+ * trimmed body is under SHORT_DECISION_MAX with no tags is a stub-capture
+ * candidate. Soft signal — capture still succeeds; the author verifies.
+ */
+function isShortDecisionNoTags(input) {
+    return (input.kind === "decision" &&
+        input.body.trim().length < exports.SHORT_DECISION_MAX &&
+        input.tags.length === 0);
+}

package/dist/mcp.js

@@ -14,6 +14,7 @@ const embed_provider_1 = require("./lib/embed-provider");
 const embed_1 = require("./lib/embed");
 const entry_1 = require("./lib/entry");
 const tags_1 = require("./lib/tags");
+const lints_1 = require("./lib/lints");
 const dedup_1 = require("./lib/dedup");
 const project_1 = require("./lib/project");
 const log_1 = require("./lib/log");
@@ -58,7 +59,7 @@ SCOPING: Retrieval is automatically scoped to the current project (resolved from
 
 OUTPUT: A markdown list of up to \`limit\` entries. Each line shows id, kind, relative age, and body. If no matching prior context exists, returns "no prior context" — treat that as signal, not an error. The entries you see are the entries your past self already thought were worth remembering; cite their ids when referencing them.`;
 async function main() {
-    const server = new index_js_1.Server({ name: "zuun", version: "0.1.1" }, { capabilities: { tools: {} } });
+    const server = new index_js_1.Server({ name: "zuun", version: "0.1.3" }, { capabilities: { tools: {} } });
     server.setRequestHandler(types_js_1.ListToolsRequestSchema, async () => ({
         tools: [
             {
@@ -140,6 +141,9 @@ async function handleRemember(args) {
                 ],
             };
         }
+        // Lint raw tags before normalization (normalizeTags lowercases; entry IDs
+        // are uppercase-hex, so an ENT token must be routed here — ENT-260514-4BA0).
+        const routing = (0, lints_1.routeEntTokensFromTags)(input.tags, input.related);
         const id = (0, id_1.newEntryId)(input.body, now);
         const entry = {
             id,
@@ -147,12 +151,23 @@ async function handleRemember(args) {
             body: input.body,
             kind: input.kind,
             source: SOURCE,
-            tags: (0, tags_1.normalizeTags)(input.tags),
-            related: input.related,
+            tags: (0, tags_1.normalizeTags)(routing.tags),
+            related: routing.related,
             stance: input.stance,
             origin: input.origin,
             project: input.project ?? (0, project_1.resolveProject)(),
         };
+        // Lint notes for the calling agent. Provenance (ENT-260514-4BA0,
+        // ENT-260514-75A3) lives in lints.ts; keep it out of agent-facing text.
+        const notes = [];
+        if (routing.routed.length > 0) {
+            notes.push(`note: moved ${routing.routed.join(", ")} from tags to related — ` +
+                `tags are for topics, related links entries`);
+        }
+        if ((0, lints_1.isShortDecisionNoTags)(entry)) {
+            notes.push(`note: this is a short decision with no tags — confirm it is a durable ` +
+                `choice, not transient state, or it reads as an accidental capture`);
+        }
         (0, entry_io_1.writeEntry)(entry);
         (0, store_1.upsertEntry)(db, entry);
         (0, log_1.appendLog)("remember", { id, kind: entry.kind, tags: entry.tags });
@@ -172,9 +187,10 @@ async function handleRemember(args) {
         })
             .catch(() => { });
         const tagLine = entry.tags.length ? ` · tags: ${entry.tags.join(", ")}` : "";
+        const noteLine = notes.length ? `\n${notes.join("\n")}` : "";
         return {
             content: [
-                { type: "text", text: `saved ${id} (${entry.kind}${tagLine})\n${entry.body}` },
+                { type: "text", text: `saved ${id} (${entry.kind}${tagLine})\n${entry.body}${noteLine}` },
             ],
         };
     }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@gorajing/zuun",
-  "version": "0.1.1",
+  "version": "0.1.3",
   "description": "Persistent memory for AI-assisted work — local-first markdown + SQLite, hybrid search, Claude Code plugin.",
   "author": "Jin Choi",
   "license": "MIT",