mirror of
https://github.com/MBombeck/HealthLog.git
synced 2026-11-09 13:01:00 +00:00
Marc-André Bombeck
502783a3ea
Replace the scalar document_id discriminator on a coach conversation with a CoachConversationDocument join table and a sticky documentScoped flag, so a conversation can be grounded on several stored documents at once while the prompt-injection fence stays intact across the data-model change. The fence is re-established at both enforcement points, fail-closed: - Client: resolveCoachSendTarget routes any turn on a fenced conversation (or a fresh chat with staged attachments) to the new /api/insights/chat/fenced endpoint; the fenced body is strict and never carries scope, guided-question, or prefill hints. - Server: the tool route fetches with documentScoped:false AND 404s when the fetched row still carries an attachment (flag/join drift), emitting a fence_drift audit alarm. Document text now enters an LLM prompt in exactly one code path — the fenced pipeline — with no tools and no health snapshot; a module-graph test pins that the tool loop is unreachable from it. The fenced turn fences each document separately with a server-built, scrubbed header line (title and filename are untrusted text), runs a per-document egress consent check on the single picked provider, and grounds numerically over the live attachment union only. Attachments are capped at five documents and 192 KB of combined text with honest truncation. Attach and detach ride dedicated endpoints: attaching validates the document is owned, live, indexed, and within cap, then flips the conversation fenced (the one legal, privilege-reducing transition, audit-logged); detaching removes the document but never clears the flag. The document-sheet chat keeps its public wire contract and shares the same fenced pipeline over the join. Migration 0245 adds the join table and the flag, backfills every legacy single-document chat into a one-attachment fenced conversation, and drops the scalar column in the same transaction — no dual-truth window.
72 lines
3.3 KiB
SQL
72 lines
3.3 KiB
SQL
-- S7: attach multiple documents to a coach conversation.
|
|
--
|
|
-- Replaces the scalar `coach_conversations.document_id` discriminator (which
|
|
-- could express only ONE document and would UNFENCE a conversation when that
|
|
-- document was deleted) with a join table + a sticky `document_scoped` flag.
|
|
--
|
|
-- The flag is the fail-closed half of the fenced-only discriminator: it is set
|
|
-- true at fenced creation / first attach and NEVER cleared. The join table
|
|
-- holds the LIVE attachment set. A count-only predicate would let a deleted
|
|
-- attachment flip a conversation (whose history may contain document-derived
|
|
-- text) back onto the tool loop; the sticky flag makes "fenced" a one-way latch
|
|
-- independent of attachment liveness.
|
|
--
|
|
-- Backfill runs in the SAME transaction as the DDL (Postgres allows it) BEFORE
|
|
-- the scalar column is dropped, so there is no dual-source-of-truth window:
|
|
-- every legacy single-doc chat becomes a one-attachment fenced conversation.
|
|
|
|
-- CreateTable
|
|
CREATE TABLE "coach_conversation_documents" (
|
|
"conversation_id" TEXT NOT NULL,
|
|
"document_id" TEXT NOT NULL,
|
|
"added_at" TIMESTAMP(3) NOT NULL DEFAULT CURRENT_TIMESTAMP,
|
|
CONSTRAINT "coach_conversation_documents_pkey" PRIMARY KEY ("conversation_id", "document_id")
|
|
);
|
|
|
|
-- CreateIndex — the document-sheet "chats about this doc" scan + the path-id
|
|
-- join filter the single-doc route uses.
|
|
CREATE INDEX "coach_conversation_documents_document_id_idx"
|
|
ON "coach_conversation_documents" ("document_id");
|
|
|
|
-- AddForeignKey — deleting a conversation cascades its attachment rows away.
|
|
ALTER TABLE "coach_conversation_documents"
|
|
ADD CONSTRAINT "coach_conversation_documents_conversation_id_fkey"
|
|
FOREIGN KEY ("conversation_id") REFERENCES "coach_conversations" ("id")
|
|
ON DELETE CASCADE ON UPDATE CASCADE;
|
|
|
|
-- AddForeignKey — deleting a document cascades its join rows away (its text
|
|
-- leaves every conversation's context) but NEVER the conversation itself.
|
|
ALTER TABLE "coach_conversation_documents"
|
|
ADD CONSTRAINT "coach_conversation_documents_document_id_fkey"
|
|
FOREIGN KEY ("document_id") REFERENCES "inbound_documents" ("id")
|
|
ON DELETE CASCADE ON UPDATE CASCADE;
|
|
|
|
-- AlterTable — the sticky fence flag.
|
|
ALTER TABLE "coach_conversations"
|
|
ADD COLUMN "document_scoped" BOOLEAN NOT NULL DEFAULT false;
|
|
|
|
-- Backfill: every legacy single-doc chat becomes a one-attachment fenced
|
|
-- conversation. Runs BEFORE the DROP COLUMN below.
|
|
INSERT INTO "coach_conversation_documents" ("conversation_id", "document_id", "added_at")
|
|
SELECT "id", "document_id", "created_at"
|
|
FROM "coach_conversations"
|
|
WHERE "document_id" IS NOT NULL;
|
|
|
|
UPDATE "coach_conversations" SET "document_scoped" = true
|
|
WHERE "document_id" IS NOT NULL;
|
|
|
|
-- The scalar discriminator goes away in the SAME migration — no dual-source-of-
|
|
-- truth window, no "in case" shim. Every reader is repointed in the same
|
|
-- release. Drop the FK + covering index first, then the column.
|
|
ALTER TABLE "coach_conversations"
|
|
DROP CONSTRAINT "coach_conversations_document_id_fkey";
|
|
|
|
DROP INDEX "coach_conversations_user_id_document_id_updated_at_idx";
|
|
|
|
ALTER TABLE "coach_conversations"
|
|
DROP COLUMN "document_id";
|
|
|
|
-- CreateIndex — the rail + tool-route-send predicate on the sticky flag.
|
|
CREATE INDEX "coach_conversations_user_id_document_scoped_updated_at_idx"
|
|
ON "coach_conversations" ("user_id", "document_scoped", "updated_at");
|