feat(linkedin-studio): SB-S1 ingest IO — writePublished/ingestText/scanInbox/listPublished [skip-docs]

Idempotent, collision-safe, create-on-demand IO under the brain dataRoot:
- writePublished: identical body → no-op; differing body at same id → disambiguated
  <id>-N.md (never clobber, never silently drop a gold record, B2 defence-in-depth).
- ingestText: provenance always published; published_date defaults to captured_at.
- scanInbox: top-level *.md only (skips .DS_Store/dotfiles/non-md), non-destructive,
  empty/absent inbox = clean no-op, re-scan re-skips via dedup.
- listPublished: parses each file in try/catch (malformed → skipped count, never
  crashes), surfaces provenance so ai-draft leakage is eyeball-visible.
9 temp-dir IO tests. brain 48→57 tests, tsc clean.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01RigJBiRFNtFZKCz21qNbQ4
This commit is contained in:
Kjell Tore Guttormsen 2026-06-23 14:54:31 +02:00
commit 0d3e4911d7
2 changed files with 277 additions and 0 deletions

View file

@ -17,6 +17,10 @@
* drops a differing body.
*/
import { existsSync, mkdirSync, readFileSync, readdirSync, writeFileSync } from "node:fs";
import { join } from "node:path";
import { dataRoot } from "./dataRoot.js";
import { mintContentId, normalizeProvenance } from "./id.js";
/** One ingested published post. `provenance` is always `published` in SB-S1. */
@ -93,3 +97,148 @@ export function parsePublishedRecord(text: string): PublishedRecord {
body,
};
}
// ── IO ───────────────────────────────────────────────────────────────────────
//
// All IO resolves through the brain package's own `dataRoot` (LINKEDIN_STUDIO_DATA
// overrides the root for tests + installs). Create-on-demand: `ingest/published/`
// is mkdir'd as needed, so ingest does NOT depend on a prior `brain init`.
const PUBLISHED_SUB = join("ingest", "published");
const INBOX_SUB = join("ingest", "inbox");
/** The body stored at `path`, or `null` if it can't be read/parsed (treat as differing). */
function storedBody(path: string): string | null {
try {
return parsePublishedRecord(readFileSync(path, "utf8")).body;
} catch {
return null;
}
}
export interface WriteResult {
/** true iff a file was written (false = identical record already present). */
written: boolean;
/** The path written or matched. */
path: string;
/** true iff an id-collision with a DIFFERING body forced a disambiguated name. */
collision?: boolean;
}
/**
* Write a record idempotently + collision-safely. If `<id>.md` already holds the
* SAME body no-op (`written:false`). If it holds a DIFFERENT body (a 48-bit hash
* collision or an unreadable file) write the next free `<id>-N.md` rather than
* clobber or silently skip so a differing gold record is never lost (B2).
*/
export function writePublished(rec: PublishedRecord): WriteResult {
const dir = dataRoot(PUBLISHED_SUB);
mkdirSync(dir, { recursive: true });
const text = serializePublishedRecord(rec);
const primary = join(dir, `${rec.id}.md`);
if (!existsSync(primary)) {
writeFileSync(primary, text, "utf8");
return { written: true, path: primary };
}
if (storedBody(primary) === rec.body) return { written: false, path: primary };
for (let n = 2; ; n++) {
const cand = join(dir, `${rec.id}-${n}.md`);
if (!existsSync(cand)) {
writeFileSync(cand, text, "utf8");
return { written: true, path: cand, collision: true };
}
if (storedBody(cand) === rec.body) return { written: false, path: cand, collision: true };
}
}
export interface IngestResult extends WriteResult {
record: PublishedRecord;
}
/**
* Mint a record from a verbatim post body and write it. `provenance` is always
* `published`; `published_date` defaults to `captured_at` when not supplied. Pure
* inputs (the caller supplies `captured_at`), idempotent + collision-safe write.
*/
export function ingestText(opts: {
body: string;
captured_at: string;
source?: string;
published_date?: string;
}): IngestResult {
const record: PublishedRecord = {
id: mintContentId(opts.body),
provenance: "published",
published_date: opts.published_date ?? opts.captured_at,
captured_at: opts.captured_at,
source: opts.source ?? "manual",
body: opts.body,
};
return { record, ...writePublished(record) };
}
export interface ScanResult {
/** ids of records written this scan. */
processed: string[];
/** ids of records already present (skipped as duplicates). */
skipped: string[];
}
/**
* Process the inbox drop-zone into `published/`. Reads TOP-LEVEL `*.md` files only,
* skipping dotfiles (`.DS_Store` etc.) and non-markdown no recursion. An empty or
* absent inbox is a clean no-op. Non-destructive in SB-S1: processed inbox files are
* left in place (a re-scan re-skips via the published dedup).
*/
export function scanInbox(opts: { captured_at: string; source?: string }): ScanResult {
const dir = dataRoot(INBOX_SUB);
const processed: string[] = [];
const skipped: string[] = [];
if (!existsSync(dir)) return { processed, skipped };
for (const entry of readdirSync(dir, { withFileTypes: true })) {
if (!entry.isFile() || entry.name.startsWith(".") || !entry.name.endsWith(".md")) continue;
const body = readFileSync(join(dir, entry.name), "utf8");
const res = ingestText({ body, captured_at: opts.captured_at, source: opts.source });
(res.written ? processed : skipped).push(res.record.id);
}
return { processed, skipped };
}
export interface PublishedSummary {
id: string;
provenance: string;
published_date: string;
firstLine: string;
}
/**
* List the published gold corpus. Each file is parsed inside a try/catch a
* malformed/hand-dropped file is counted as `skipped`, never crashing the command.
* Surfaces `provenance` so the operator can eyeball that nothing `ai-draft` leaked.
*/
export function listPublished(): { records: PublishedSummary[]; skipped: number } {
const dir = dataRoot(PUBLISHED_SUB);
const records: PublishedSummary[] = [];
let skipped = 0;
if (!existsSync(dir)) return { records, skipped };
for (const name of readdirSync(dir)) {
if (name.startsWith(".") || !name.endsWith(".md")) continue;
try {
const rec = parsePublishedRecord(readFileSync(join(dir, name), "utf8"));
records.push({
id: rec.id,
provenance: rec.provenance,
published_date: rec.published_date,
firstLine: rec.body.split("\n", 1)[0],
});
} catch {
skipped++;
}
}
records.sort((a, b) => a.published_date.localeCompare(b.published_date));
return { records, skipped };
}

View file

@ -0,0 +1,128 @@
import { describe, test, beforeEach, afterEach } from "node:test";
import assert from "node:assert/strict";
import { mkdtempSync, rmSync, existsSync, readFileSync, writeFileSync, mkdirSync, readdirSync } from "node:fs";
import { join } from "node:path";
import { tmpdir } from "node:os";
import {
writePublished,
ingestText,
scanInbox,
listPublished,
serializePublishedRecord,
parsePublishedRecord,
type PublishedRecord,
} from "../src/ingest.js";
import { mintContentId } from "../src/id.js";
const CAPTURED = "2026-06-23";
function publishedDir(root: string): string {
return join(root, "ingest", "published");
}
function inboxDir(root: string): string {
return join(root, "ingest", "inbox");
}
function mdFiles(dir: string): string[] {
return existsSync(dir) ? readdirSync(dir).filter((f) => f.endsWith(".md")) : [];
}
describe("ingest IO — writePublished / ingestText / scanInbox / listPublished (SC1/SC3/SC4)", () => {
let root: string;
const prevEnv = process.env.LINKEDIN_STUDIO_DATA;
beforeEach(() => {
root = mkdtempSync(join(tmpdir(), "brain-ingest-"));
process.env.LINKEDIN_STUDIO_DATA = root;
});
afterEach(() => {
if (prevEnv === undefined) delete process.env.LINKEDIN_STUDIO_DATA;
else process.env.LINKEDIN_STUDIO_DATA = prevEnv;
rmSync(root, { recursive: true, force: true });
});
test("create-on-demand: ingestText writes ingest/published/<id>.md with no prior init (SC1)", () => {
const body = "Min første publiserte post.\n\nMed avsnitt.";
const res = ingestText({ body, captured_at: CAPTURED });
assert.equal(res.written, true);
const path = join(publishedDir(root), `${mintContentId(body)}.md`);
assert.ok(existsSync(path), "record file created at runtime data-path");
const rec = parsePublishedRecord(readFileSync(path, "utf8"));
assert.equal(rec.provenance, "published");
assert.equal(rec.body, body, "verbatim body preserved");
});
test("published_date defaults to captured_at when not supplied", () => {
const res = ingestText({ body: "no date given", captured_at: CAPTURED });
assert.equal(res.record.published_date, CAPTURED);
});
test("published_date is kept when supplied", () => {
const res = ingestText({ body: "dated", captured_at: CAPTURED, published_date: "2026-05-26" });
assert.equal(res.record.published_date, "2026-05-26");
});
test("re-ingesting an identical body is a no-op (SC3 dedup): {written:false}, no duplicate file", () => {
const body = "same exact text";
ingestText({ body, captured_at: CAPTURED });
const res2 = ingestText({ body, captured_at: "2026-07-01" }); // different day, same body
assert.equal(res2.written, false);
assert.equal(mdFiles(publishedDir(root)).length, 1, "still exactly one record");
});
test("id-collision with a DIFFERING body disambiguates to <id>-2.md — never silent data loss (B2)", () => {
const body = "the real body";
const id = mintContentId(body);
// Pre-plant a file at <id>.md whose body is DIFFERENT (simulates a hash collision).
mkdirSync(publishedDir(root), { recursive: true });
const squatter: PublishedRecord = {
id, provenance: "published", published_date: CAPTURED, captured_at: CAPTURED,
source: "manual", body: "a DIFFERENT body that happens to share the id",
};
writeFileSync(join(publishedDir(root), `${id}.md`), serializePublishedRecord(squatter), "utf8");
const res = writePublished({ ...squatter, body });
assert.equal(res.written, true);
assert.equal(res.collision, true);
assert.match(res.path, /-2\.md$/);
assert.equal(mdFiles(publishedDir(root)).length, 2, "both records kept — no data loss");
});
test("scanInbox processes top-level *.md only, skipping dotfiles like .DS_Store (SC4)", () => {
mkdirSync(inboxDir(root), { recursive: true });
writeFileSync(join(inboxDir(root), "post-a.md"), "Post A body", "utf8");
writeFileSync(join(inboxDir(root), "post-b.md"), "Post B body", "utf8");
writeFileSync(join(inboxDir(root), ".DS_Store"), "junk", "utf8");
writeFileSync(join(inboxDir(root), "notes.txt"), "not markdown", "utf8");
const res = scanInbox({ captured_at: CAPTURED });
assert.equal(res.processed.length, 2, "only the two .md files processed");
assert.equal(mdFiles(publishedDir(root)).length, 2);
});
test("scanInbox on an empty/absent inbox is a clean no-op (SC4)", () => {
const res = scanInbox({ captured_at: CAPTURED });
assert.deepEqual(res, { processed: [], skipped: [] });
});
test("re-running scanInbox re-skips already-published records (idempotent)", () => {
mkdirSync(inboxDir(root), { recursive: true });
writeFileSync(join(inboxDir(root), "post.md"), "once", "utf8");
scanInbox({ captured_at: CAPTURED });
const res2 = scanInbox({ captured_at: CAPTURED });
assert.equal(res2.processed.length, 0);
assert.equal(res2.skipped.length, 1);
assert.equal(mdFiles(publishedDir(root)).length, 1, "no duplicate");
});
test("listPublished surfaces provenance and skips a malformed file without throwing", () => {
ingestText({ body: "good record", captured_at: CAPTURED });
mkdirSync(publishedDir(root), { recursive: true });
writeFileSync(join(publishedDir(root), "broken.md"), "not a valid record at all", "utf8");
const res = listPublished();
assert.equal(res.records.length, 1, "the one valid record listed");
assert.equal(res.records[0].provenance, "published");
assert.equal(res.skipped, 1, "the malformed file counted as skipped, not a crash");
});
});