feat(core): add pi event adapter for 0.41 ingest by ZeR020 · Pull Request #1546 · kunickiaj/codemem

ZeR020 · 2026-08-29T07:40:43Z

Second PR in the pi integration stack — context and overall design in #1430. Builds on #1473 (merged); the diff here is the pi adapter only.

Why

Pi events need a core adapter that produces the same envelope shape ingestRawEvents already accepts, with source: "pi" explicit and deterministic ids so HTTP/CLI retries do not duplicate rows.

What

packages/core/src/pi-hooks.ts: map session/prompt/tool events; compaction is flush-only
Envelope fields include session_stream_id / session_id aliases required by 0.41 ingest
Test: ingestRawEvents persists source pi and creates zero opencode rows

Test plan

vitest run packages/core/src/pi-hooks.test.ts — 40/40 passed
Gates lint, tsc, build green locally; test passes except local-sandbox env failures unrelated to this change (sqlite-vec native ext unavailable in this dev sandbox)

@kunickiaj layer 2 of the stack is ready for review whenever you have a moment. Layers 3–8 will follow one at a time as each predecessor merges, same as this one.

Related: #1430, #1429, #1473

Map pi extension events to AdapterEvent envelopes with source "pi" and deterministic ids so later stacked routes can call ingestRawEvents without a parallel writer.

Copilot

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4e8d1941eb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-08-29T07:44:29Z

+		const role = coerceString(field(payload, "role")).toLowerCase();
+		const text = coerceString(field(payload, "text", "content", "prompt"));
+		if (!text) return null;
+		if (!entryId) return null;


Read message_end fields from event.message

When this adapter receives Pi 0.41's native message_end event, the role, content, and message id are nested under event.message, not provided as top-level role, text, and entryId fields. This branch therefore sees an empty role and returns null for every real user and assistant message, so prompts and assistant transcript content never reach the raw-event sweeper. Unwrap the nested message (including text content/blocks and its id) before applying this mapping.

Useful? React with 👍 / 👎.

Correct about Pi 0.41's native message_end shape — but this adapter intentionally never sees the raw pi event. The pi extension's pi.on("message_end") handler unwraps event.message (role/text via extractMessageRole/extractMessageText) and synthesizes a deterministic entryId (stableMessageEntryId, timestamp-discriminated) before POSTing the flat payload this mapper consumes (packages/pi-extension/src/index.ts, payloads.ts — landing in a later layer of this stack, see #1430). Keeping the unwrap on the producer side is what lets core stay pi-agnostic. Traced end-to-end: no caller passes the raw nested event. No change needed here.

kunickiaj

Two ingestion-identity issues need correction before this adapter is safe to merge.

kunickiaj · 2026-08-30T02:42:11Z

+ */
+function buildPiEventId(sessionId: string, stablePart: string): string {
+	const part = stablePart.trim();
+	if (!part) throw new Error("pi event id stable part is required");


Raw ingestion validates event_id against /^[A-Za-z0-9._:-]+$/ and a 128-character limit, but this concatenates unconstrained session and tool/message IDs. For example, sessionId="session/with/slash" and entryId="entry+1" produces an ID that ingestRawEvents rejects with RawEventIngestValidationError. Pi toolCallId is typed only as string, so provider-specific IDs can trigger this even if session IDs are UUIDs. Please hash the identity components into a fixed-length safe ID, as the Claude and Codex adapters do, and add charset/length regression coverage.

Hashed identity into pi_evt_ + sha256(sessionId|piEvent|idPart).hex.slice(0, 24), matching Claude/Codex. Added charset/length coverage, including session/with/slash + entry+1.

05d7b2f

kunickiaj · 2026-08-30T02:42:11Z

+		const reason = field(payload, "reason");
+		eventType = "session_end";
+		eventPayload = { reason: reason ?? null };
+		idPart = entryId || "session_end";


This fallback gives every shutdown for a session the same event ID. Pi can shut an extension down for reload and later shut the same session down for exit; the second event is then deduplicated, preserving the earlier reason/timestamp and losing the final lifecycle event. The continuation stack's payload builder also currently supplies the fixed entryId "session_end", so it hits the same collision. Please require or derive a retry-stable discriminator for each distinct shutdown occurrence and add a reload-then-exit regression test.

Shutdown no longer uses a fixed session_end id. When entryId is missing or that generic value, the discriminator is session_end:<reason>. Reload then quit persist as two events; identical retries still dedupe.

05d7b2f

Raw ingest rejects concatenated session/entry ids that contain slash or plus. Hash identity components like Claude/Codex. session_shutdown fell back to a fixed session_end id, so reload then quit collapsed to one row. Derive idPart from reason when entryId is missing or the generic session_end.

ZeR020 · 2026-08-30T12:40:04Z

@kunickiaj requested identity fixes are in 05d7b2f7 (hashed event IDs + shutdown discriminator). CI is green. Ready for re-review.

ZeR020 · 2026-09-02T09:50:19Z

@kunickiaj ready for your review whenever you have a moment.

kunickiaj

Thanks for addressing the original review. The hashed-ID fix looks good. One shutdown edge case still blocks approval; the other two comments are small identity-contract follow-ups that should fit naturally with that change.

kunickiaj · 2026-09-02T23:52:23Z

+		const reason = field(payload, "reason");
+		eventType = "session_end";
+		eventPayload = { reason: reason ?? null };
+		idPart = entryId && entryId !== "session_end" ? entryId : `session_end:${coerceString(reason)}`;


Two shutdowns with the same reason (for example, /reload twice) still produce the same ID, so the second is dropped. Please have the producer send a stable per-occurrence entryId—or use its stable payload timestamp as the fallback—and add reload → reload coverage.

kunickiaj · 2026-09-02T23:52:23Z

+
+	if (!idPart) return null;
+
+	const meta: Record<string, unknown> = {


Please match the Claude and Codex adapters by exporting PI_EVENT_ID_ALGO = "pi/1" and storing it as meta.event_id_algo. This keeps stored IDs attributable if the algorithm changes.

kunickiaj · 2026-09-02T23:52:23Z

+		expect(result.skipped).toBe(0);
+	});
+
+	it("persists reload then exit as distinct session_end events", () => {


This covers reload → quit, whose reasons already differ. Please add reload → reload: both occurrences should insert, while retrying either payload should skip.

feat(core): add pi event adapter for 0.41 ingest

4e8d194

Map pi extension events to AdapterEvent envelopes with source "pi" and deterministic ids so later stacked routes can call ingestRawEvents without a parallel writer.

Copilot AI lite review requested due to automatic review settings August 29, 2026 07:40

ZeR020 mentioned this pull request Aug 29, 2026

feat(core): add pi event adapter for 0.41 ingest ZeR020/codemem#1

Closed

Copilot AI reviewed Aug 29, 2026

chatgpt-codex-connector Bot reviewed Aug 29, 2026

View reviewed changes

kunickiaj requested changes Aug 30, 2026

View reviewed changes

Merge remote-tracking branch 'origin/main' into stack/2-pi-hooks-adapter

fe16e4a

kunickiaj requested changes Sep 2, 2026

View reviewed changes


		if (!idPart) return null;

		const meta: Record<string, unknown> = {

Conversation

ZeR020 commented Aug 29, 2026

Why

What

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Aug 29, 2026

Choose a reason for hiding this comment

Uh oh!

ZeR020 Aug 29, 2026

Choose a reason for hiding this comment

Uh oh!

kunickiaj left a comment

Choose a reason for hiding this comment

Uh oh!

kunickiaj Aug 30, 2026

Choose a reason for hiding this comment

Uh oh!

ZeR020 Aug 30, 2026

Choose a reason for hiding this comment

Uh oh!

kunickiaj Aug 30, 2026

Choose a reason for hiding this comment

Uh oh!

ZeR020 Aug 30, 2026

Choose a reason for hiding this comment

Uh oh!

ZeR020 commented Aug 30, 2026

Uh oh!

ZeR020 commented Sep 2, 2026

Uh oh!

kunickiaj left a comment

Choose a reason for hiding this comment

Uh oh!

kunickiaj Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kunickiaj Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

kunickiaj Sep 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

kunickiaj Sep 2, 2026 •

edited

Loading

kunickiaj Sep 2, 2026 •

edited

Loading

kunickiaj Sep 2, 2026 •

edited

Loading