2.6.C — the reseat no longer blanks the screen, and /cost shows where the money went

[cemililik]

Closes 2.6.C. Two maintainer manual-test findings, and a residual that turned out to need its own ADR.

F1 — a /models reseat blanked the screen

A mid-session model switch builds a brand-new view store. On the full-screen renderer — the TTY default since 2.6.F, and whose alt buffer has no native scrollback behind it — that store is the scrollback. It was seeded transcript: [], so the entire conversation vanished the moment a user switched models.

The durable transcript and the new model's context were never at risk. The loss was the view — which is exactly why it slipped through: the existing reseat test is literally named "carrying the transcript" and asserted turnCount, model, sessionId and DB rows. Never state.transcript.

The reseat now carries the rendered conversation, and the notice lands beneath it as an inline marker:

⇄ model changed claude-sonnet-4-6 → claude-opus-4-8 — the new model sees the text
  transcript only (not prior tool calls or file contents).

The carry is gated to full-screen, and that gate is load-bearing in both directions. driveInk re-mounts ink per session, which resets <Static>'s index — so a seeded inline store would re-print the whole conversation on top of the copy the terminal's scrollback already holds. Inline needs no carry; its lines survive the swap. /clear carries nothing on either renderer — its rebuild path has no carry parameter at all, so its freshness is structural.

TranscriptBound became a closed literal union with no defaults, because the dangerous axis was never the seed — it was the bound. A full-screen caller that forgot the argument used to fall back to inline silently: F1 again, with a green build. Now it is a compile error.

/cost — where the money went

ADR-0059 planned to derive the breakdown from the per-message model_id. That cannot work. One row holds one model, but a turn whose tool loop fails over bills two — pinned by an existing core test. So this needed its own decision: ADR-0070.

A durable session_costs table, with the invariant

SUM(session_costs.cost_microcents) == agent_sessions.total_cost_microcents

true by construction — both sides fed by the same cost:updated egress, same arithmetic, same transaction, single owner. /cost is a money surface: a breakdown whose rows do not sum to the total shown is worse than no breakdown.

It closed a live bug on the way: total_cost_microcents was being SET blindly by five writers, so two chat-resume processes on one session could permanently corrupt the total. It now has exactly one.

Two honesties the panel carries: a $0 row with real tokens is a model we could not price, not a free one (2.6.Q's F5 is precisely that unpriced models are where the cost cap silently stops applying); and a session older than 2.6.C says its breakdown is unavailable rather than pretending it used one model.

Stated plainly, not buried: the guarantee is against Relavium's own event stream, not the provider's invoice. An egress that streamed content but ended without a usage chunk, and a mid-stream failure, are recorded as 0 on both sides — so reported spend is a systematic under-estimate. That is a usage-capture gap in the seam, filed against 2.6.Q.

What the reviews caught in my own work

Four adversarial rounds ran over this. Each found something real, and the record is worth having:

• A tripwire I shipped was armed in production (nothing sets NODE_ENV; apps/cli has no error boundary) — and ink swallowed it anyway (no-op error callbacks), so it could never fail a test. Probed and confirmed. It moved to the composition roots, where a throw actually escapes.
• My own fixture rewrite turned two tests into vacuous passes — they asserted over a chat that was no longer rendered.
• A perf harness measured a pointer assignment (the seed aliases the array), while I had amended ADR-0059 to say it verified an O(n) claim. Strictly worse than citing nothing. It now measures wrapTranscript, and a quadratic regression fails it.
• The /cost DB read — the step's central change — had zero test coverage at any layer. Reverting it to memory left the suite green.

All folded, each with a break-verified regression.

ADR notes (append-only)

• ADR-0068 gains the limitation its own defaults created: DEFAULT_ALT_SCREEN=true plus Decision (c)'s move of the transcript into the in-memory viewport is what turned the ADR-0059 reseat into a visible loss.
• ADR-0059 gains three corrections — including that the perf harness it claimed was "verified by the 2.6.C harness" never existed. It does now.

Verification

pnpm turbo run lint typecheck test build — 24/24 green. 2132 CLI · 985 core · 207 db · 543 llm · 459 shared.

Every fix is break-verified: the bug is re-introduced, the test is shown to fail, the code is restored. F1's break reproduces the user's screen exactly — expected [ 'notice' ] to deeply equal [ 'user', 'assistant', 'notice' ].

Migration 0009 creates session_costs, drops session_messages' never-written token/cost columns, and backfills a (pre-2.6.C) sentinel row per legacy session so the invariant holds for every row — not just post-migration ones.

🤖 Generated with Claude Code

Summary by CodeRabbit

• New Features
  • /cost now shows durable per-session totals plus a per-model breakdown with token usage, call counts, and “price unknown” indicators.
  • /models reseat includes an explicit old → new notice and (in full-screen) preserves the rendered conversation.
• Bug Fixes
  • /cost after resume reflects previously persisted spend breakdown (and no longer shows transient zeros).
  • Improved priced/unpriced/errored-turn reconciliation and model-id output sanitization.
  • Added UI consistency checks to prevent inline vs full-screen transcript viewport mismatches.
• Documentation
  • Updated CLI and reference docs for the new /cost output, reseat behavior, and accessibility wording.

[sourcery-ai]

Sorry @cemililik, your pull request is larger than the review limit of 150000 diff characters

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

📝 Walkthrough

Walkthrough

This change adds durable per-model session cost attribution, database-backed /cost breakdowns, renderer-aware transcript carry across /models reseats, explicit transcript-bound contracts, renderer/store guards, migrations, tests, and documentation updates.

Changes

Session costs and renderer-aware reseat

Layer / File(s)	Summary
Durable cost persistence and reconciliation packages/db/..., packages/shared/src/run-event.ts, apps/cli/src/chat/persister.ts	cost:updated events persist additive per-model rows and session totals transactionally, with priced/unpriced tracking and legacy backfill support.
Cost display and CLI integration apps/cli/src/chat/repl-info.ts, apps/cli/src/commands/chat.ts, apps/cli/src/chat/*test.ts	/cost loads durable totals and renders sanitized per-model rows; model-switch notices describe the old-to-new transition and carried content.
Transcript-bound contract and reseat wiring apps/cli/src/render/tui/session-view-model.ts, apps/cli/src/render/tui/chat-store.ts, apps/cli/src/commands/chat.ts, apps/cli/src/home/drive-home.tsx, apps/cli/src/render/tui/chat-ink.tsx, apps/cli/src/render/tui/home-controller.ts	Explicit inline/full-screen bounds control transcript seeding, reseat callbacks carry rendered entries, and runtime guards enforce renderer/store agreement.
Validation, migrations, and references apps/cli/**/*test.ts, apps/cli/src/render/tui/*test.tsx, packages/db/drizzle/*, packages/db/src/*test.ts, docs/...	Tests cover cost reconciliation, renderer behavior, reseat carry, /clear, resume display, migrations, and bound invariants; ADRs and references document the resulting contracts.

Estimated code review effort: 5 (Critical) | ~120 minutes

Sequence Diagram(s)

sequenceDiagram
  participant User
  participant ChatREPL
  participant SessionPersister
  participant SessionStore
  participant Renderer
  User->>ChatREPL: /models or /cost
  ChatREPL->>Renderer: capture rendered transcript for reseat
  ChatREPL->>SessionPersister: process model reseat and cost events
  SessionPersister->>SessionStore: persist per-model cost attribution
  ChatREPL->>Renderer: seed full-screen transcript or render cost breakdown

Loading

Possibly related PRs

• HodeTech/Relavium#2: Shares the CostUpdatedEvent contract used by durable cost attribution.
• HodeTech/Relavium#54: Shares the cost:updated persister path extended here.
• HodeTech/Relavium#74: Shares the renderer transcript-bound and full-screen TUI wiring extended here.

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 77.42% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly summarizes the two main changes: fixing reseat screen blanking and improving /cost breakdown visibility.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch development

---

_{Comment @coderabbitai help to get the list of available commands.}

[gemini-code-assist]

Code Review

This pull request implements durable per-model session cost attribution (ADR-0070) by introducing a new session_costs table and dropping unused token and cost columns from session_messages. It updates the /cost command to read the breakdown from the database, ensuring accuracy across resumed sessions, and adds extensive tests for the per-model cost breakdown and the transcript carry on alt-screen. The review feedback correctly identifies critical transaction isolation issues in packages/db/src/session-store.ts, where database operations inside db.transaction callbacks are incorrectly executed on the outer db instance instead of the transaction context parameter.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

In Drizzle ORM, queries executed inside a transaction must use the transaction context parameter (usually named 'tx') passed to the callback. Executing queries on the outer 'db' instance runs them outside the transaction, which breaks the atomicity and isolation guarantees of the transaction. Use the 'tx' parameter to execute the insert and update queries within the transaction.

      withBusyRetry(() =>
        db.transaction(
          (tx) => {
            tx.insert(sessionCosts)
              .values({
                id: entry.id,
                sessionId: entry.sessionId,
                model: entry.model,
                modelCatalogId: entry.modelCatalogId ?? null,
                inputTokens: entry.inputTokens,
                outputTokens: entry.outputTokens,
                costMicrocents: entry.costMicrocents,
                callCount: 1,
                unpricedCalls: entry.priced ? 0 : 1,
                createdAt: entry.ts,
                updatedAt: entry.ts,
              })
              .onConflictDoUpdate({
                target: [sessionCosts.sessionId, sessionCosts.model],
                set: {
                  inputTokens: sql.raw("input_tokens + " + entry.inputTokens),
                  outputTokens: sql.raw("output_tokens + " + entry.outputTokens),
                  costMicrocents: sql.raw("cost_microcents + " + entry.costMicrocents),
                  callCount: sql.raw("call_count + 1"),
                  unpricedCalls: sql.raw("unpriced_calls + " + (entry.priced ? 0 : 1)),
                  updatedAt: entry.ts,
                },
              })
              .run();
            tx.update(agentSessions)
              .set({
                totalCostMicrocents: sql.raw("total_cost_microcents + " + entry.costMicrocents),
                updatedAt: entry.ts,
              })
              .where(eq(agentSessions.id, entry.sessionId))
              .run();
          },
          { behavior: 'immediate' },
        ),
      );

[gemini-code-assist]

Inside 'loadSessionCostBreakdown', the queries 'loadSession' and 'readCostRows' are executed on the outer 'db' instance instead of the transaction context 'tx'. This defeats the snapshot isolation guarantee of the transaction. Query the database directly using 'tx' to ensure both reads occur within the same transaction snapshot. Additionally, when constructing the returned rows, use the conditional spread pattern for the optional 'modelCatalogId' property to maintain codebase consistency and avoid setting explicit 'undefined' values under 'exactOptionalPropertyTypes'.

      withBusyRetry(() =>
        db.transaction((tx) => {
          const row = tx
            .select()
            .from(agentSessions)
            .where(eq(agentSessions.id, sessionId))
            .get();
          const totalCostMicrocents = row?.totalCostMicrocents ?? 0;
          const rows = tx
            .select()
            .from(sessionCosts)
            .where(eq(sessionCosts.sessionId, sessionId))
            .orderBy(desc(sessionCosts.costMicrocents))
            .all()
            .map((r) => ({
              model: r.model,
              ...(r.modelCatalogId === null || r.modelCatalogId === undefined ? {} : { modelCatalogId: r.modelCatalogId }),
              inputTokens: r.inputTokens,
              outputTokens: r.outputTokens,
              costMicrocents: r.costMicrocents,
              callCount: r.callCount,
              unpricedCalls: r.unpricedCalls,
            }));
          return { totalCostMicrocents, rows };
        }),
      )

References

1. Prefer the conditional spread pattern ...(x === undefined ? {} : { x }) (or with null checks) for optional properties in TypeScript to maintain codebase consistency and conciseness. Avoid replacing it with verbose helper objects with multiple if statements or lateral rest-destructuring style changes, as the conditional spread is correct under exactOptionalPropertyTypes and prevents explicit undefined values from reaching the target.

[coderabbitai]

Actionable comments posted: 8

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@apps/cli/src/render/tui/chat-projection.test.ts`:
- Around line 1097-1117: Update the performance test around medianWrapMs so each
benchmark sample measures cold wrapping by supplying fresh TranscriptEntry
objects, or otherwise clear entryWrapCache before every sample. Keep the
existing 200-message assertion and budget check, and adjust the warm-up/comment
wording only if the test intentionally measures steady-state cached latency
instead.

In `@apps/cli/src/render/tui/home-app.test.tsx`:
- Around line 940-945: Update the assertions in the conversation-switch test
around the seeded `store.notice('assistant: four')` entry to verify that the
carried assistant output is present and preserved, rather than only checking
user messages and the switch marker. Apply the same assertion change to the
corresponding test block near the second referenced section.

In `@docs/analysis/models-dev-dynamic-catalog-enrichment-2026-07-11.md`:
- Around line 179-181: Specify the fenced block language for the precedence
string by labeling the Markdown fence as text, preserving the existing content
unchanged.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md`:
- Line 71: Update the documented recordSessionCost signature to include the
required id field alongside sessionId and the other entry properties. Keep the
signature consistent with SessionStore.recordSessionCost and the ADR’s statement
that row IDs are generated via deps.uuid().

In `@docs/reference/contracts/sse-event-schema.md`:
- Line 81: Add the optional priced?: boolean property to the CostUpdatedEvent
interface alongside the existing cost and token fields, matching the documented
cost:updated event shape and preserving backward compatibility for readers and
producers that omit it.

In `@docs/reference/desktop/database-schema.md`:
- Around line 392-398: Update the later mapping note around the per-message
column list to remove input_tokens, output_tokens, and cost_microcents, leaving
only columns that still exist such as model_id. Direct durable cost attribution
to session_costs, consistent with the migration 0009 documentation.

In `@docs/roadmap/phases/phase-2.6-conversational-authoring.md`:
- Around line 220-222: Update the blockquote around the “Status” note by
prefixing the blank line between the consecutive quoted lines with “>” or
merging the notes, preserving the content while making the blockquote continuous
and satisfying MD028.

In `@docs/tutorials/cli/start-a-chat-session.md`:
- Around line 51-55: Update the model-switch sentence in the tutorial text to
use the complete English phrasing “The model can be changed mid-chat with
/models,” while preserving the surrounding explanation and formatting.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 9ee68cbe-fbec-48e1-92d7-208101018ebd

📥 Commits

Reviewing files that changed from the base of the PR and between 88d43d9 and b012af4.

📒 Files selected for processing (48)

• apps/cli/src/chat/persister.test.ts
• apps/cli/src/chat/persister.ts
• apps/cli/src/chat/repl-info.test.ts
• apps/cli/src/chat/repl-info.ts
• apps/cli/src/chat/session-host.test.ts
• apps/cli/src/commands/chat.test.ts
• apps/cli/src/commands/chat.ts
• apps/cli/src/home/drive-home.test.ts
• apps/cli/src/home/drive-home.tsx
• apps/cli/src/render/tui/chat-app.test.tsx
• apps/cli/src/render/tui/chat-ink.tsx
• apps/cli/src/render/tui/chat-projection.test.ts
• apps/cli/src/render/tui/chat-store.test.ts
• apps/cli/src/render/tui/chat-store.ts
• apps/cli/src/render/tui/home-app.test.tsx
• apps/cli/src/render/tui/home-controller.test.ts
• apps/cli/src/render/tui/home-controller.ts
• apps/cli/src/render/tui/session-view-model.test.ts
• apps/cli/src/render/tui/session-view-model.ts
• docs/analysis/README.md
• docs/analysis/models-dev-dynamic-catalog-enrichment-2026-07-11.md
• docs/architecture/agent-sessions.md
• docs/decisions/0059-cli-mid-session-model-reseat.md
• docs/decisions/0068-full-screen-tui-renderer-ink7-harness.md
• docs/decisions/0069-string-width-for-the-cli-renderer.md
• docs/decisions/0070-durable-per-model-session-cost-attribution.md
• docs/decisions/README.md
• docs/glossary.md
• docs/reference/cli/accessibility.md
• docs/reference/cli/chat-session.md
• docs/reference/cli/home.md
• docs/reference/contracts/agent-session-spec.md
• docs/reference/contracts/sse-event-schema.md
• docs/reference/desktop/database-schema.md
• docs/reference/vscode/extension-api.md
• docs/roadmap/current.md
• docs/roadmap/deferred-tasks.md
• docs/roadmap/phases/phase-2.6-conversational-authoring.md
• docs/tutorials/cli/start-a-chat-session.md
• packages/core/src/engine/agent-turn.ts
• packages/db/drizzle/0009_lonely_marauders.sql
• packages/db/drizzle/meta/0009_snapshot.json
• packages/db/drizzle/meta/_journal.json
• packages/db/src/index.ts
• packages/db/src/schema.ts
• packages/db/src/session-store.test.ts
• packages/db/src/session-store.ts
• packages/shared/src/run-event.ts

[coderabbitai]

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (3)

docs/decisions/0070-durable-per-model-session-cost-attribution.md (3)

52-53: 🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win

Define call_count consistently as all cost events.

The write-path description increments call_count for every cost:updated, including unpriced calls, while the schema comment says “billed egresses.” Since /cost reports unpriced calls as N of M, M must include those calls. Update the contract wording to avoid inconsistent implementations.

Also applies to: 71-75, 172-175

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` around
lines 52 - 53, Update the contract wording for call_count in the schema comment
and the referenced write-path sections to define it as the total number of cost
events, including unpriced calls. Ensure the /cost N-of-M semantics use
call_count as the full event count, while unpriced_calls remains the subset that
could not be priced.

---

132-139: 🗄️ Data Integrity & Integration | 🟠 Major | 🏗️ Heavy lift

Use a collision-proof legacy-session marker.

cost:updated.model is a raw string and supports custom/self-hosted models; the event contract does not forbid parentheses. A real model named (pre-2.6.C) would collide with the backfill row’s (session_id, model) key and merge new spend into the “breakdown unavailable” bucket. Reserve the sentinel through schema validation or add an explicit legacy-row discriminator.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` around
lines 132 - 139, Revise the legacy backfill design described in the ADR to
prevent collisions with custom model identifiers: reserve the legacy-session
marker through schema validation, or add an explicit discriminator that
separates legacy aggregate rows from real model rows. Update the invariant and
/cost rendering description to use the selected collision-proof representation,
and ensure new spend cannot merge into the legacy bucket.

---

41-67: 📐 Maintainability & Code Quality | 🟠 Major | ⚡ Quick win

Remove or complete the duplicated schema block.

This ADR identifies database-schema.md as the canonical schema home, but then restates the full CREATE TABLE and indexes. The duplicate also omits the CHECK (model <> '') claimed at Line 67. Keep only the decision-level constraints and link to the canonical schema, or make the example complete and explicitly illustrative.

As per coding guidelines, specifications must have one canonical home under docs/reference/; link to them rather than restating them.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` around
lines 41 - 67, Remove the duplicated CREATE TABLE and index SQL block from the
ADR, retaining only the decision-level constraints and rationale. Update the
surrounding text to link to the canonical session_costs definition in
database-schema.md, including the model non-empty CHECK constraint, without
restating the full schema.

Source: Coding guidelines

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Outside diff comments:
In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md`:
- Around line 52-53: Update the contract wording for call_count in the schema
comment and the referenced write-path sections to define it as the total number
of cost events, including unpriced calls. Ensure the /cost N-of-M semantics use
call_count as the full event count, while unpriced_calls remains the subset that
could not be priced.
- Around line 132-139: Revise the legacy backfill design described in the ADR to
prevent collisions with custom model identifiers: reserve the legacy-session
marker through schema validation, or add an explicit discriminator that
separates legacy aggregate rows from real model rows. Update the invariant and
/cost rendering description to use the selected collision-proof representation,
and ensure new spend cannot merge into the legacy bucket.
- Around line 41-67: Remove the duplicated CREATE TABLE and index SQL block from
the ADR, retaining only the decision-level constraints and rationale. Update the
surrounding text to link to the canonical session_costs definition in
database-schema.md, including the model non-empty CHECK constraint, without
restating the full schema.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 7a804b4a-a839-4d2c-bdd7-51e7f860e77e

📥 Commits

Reviewing files that changed from the base of the PR and between b012af4 and 38d556b.

📒 Files selected for processing (10)

• apps/cli/src/render/tui/chat-projection.test.ts
• apps/cli/src/render/tui/home-app.test.tsx
• apps/cli/src/render/tui/session-view-model.ts
• docs/analysis/models-dev-dynamic-catalog-enrichment-2026-07-11.md
• docs/decisions/0070-durable-per-model-session-cost-attribution.md
• docs/reference/contracts/sse-event-schema.md
• docs/reference/desktop/database-schema.md
• docs/roadmap/phases/phase-2.6-conversational-authoring.md
• docs/tutorials/cli/start-a-chat-session.md
• packages/db/src/session-store.ts

🚧 Files skipped from review as they are similar to previous changes (7)

• docs/tutorials/cli/start-a-chat-session.md
• docs/analysis/models-dev-dynamic-catalog-enrichment-2026-07-11.md
• apps/cli/src/render/tui/chat-projection.test.ts
• docs/reference/desktop/database-schema.md
• apps/cli/src/render/tui/session-view-model.ts
• docs/roadmap/phases/phase-2.6-conversational-authoring.md
• apps/cli/src/render/tui/home-app.test.tsx

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (4)

docs/decisions/0070-durable-per-model-session-cost-attribution.md (3)

60-62: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win

Document the is_legacy discriminator in the upsert target.

This section says the upsert targets (session_id, model), but the schema and implementation use (session_id, model, is_legacy). Following this text would reintroduce the sentinel/custom-model collision that this ADR is intended to prevent.

Suggested fix

-1. an **upsert** on `(session_id, model)` that
+1. an **upsert** on `(session_id, model, is_legacy)` that

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` around
lines 60 - 62, Update the upsert target description in the ADR to include the
is_legacy discriminator, specifying (session_id, model, is_legacy) rather than
only (session_id, model). Preserve the additive update semantics and explain
that the discriminator keeps sentinel and custom-model records distinct.

---

81-81: 🗄️ Data Integrity & Integration | 🟡 Minor | ⚡ Quick win

Define the invariant for zero-spend sessions.

A fresh session has no session_costs rows, so SQL SUM(...) returns NULL, not 0. As written, the invariant is false for those sessions. State that an empty sum is zero or express it with COALESCE(SUM(...), 0).

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` at line
81, Clarify the invariant for every agent_sessions row so sessions without
session_costs entries treat the empty sum as zero. Update the displayed SUM
expression to use COALESCE(SUM(...), 0), or explicitly state that an empty sum
equals zero, while preserving the existing equality to total_cost_microcents.

---

167-172: 🗄️ Data Integrity & Integration | 🟠 Major | ⚡ Quick win

Document the transactional /cost read API.

The store provides loadSessionCostBreakdown(sessionId) to return the total and rows from one transaction, but this ADR names only loadSessionCosts(sessionId). That omits the API needed to prevent the panel from combining rows and totals from different snapshots.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md` around
lines 167 - 172, Update the `/cost` read API description to name
`SessionStore.loadSessionCostBreakdown(sessionId)` as the transactional source
for the total and per-model rows, replacing or clarifying the standalone
`loadSessionCosts(sessionId)` reference. State that the panel must obtain both
values from this single transaction snapshot rather than combining separate
reads.

apps/cli/src/chat/repl-info.ts (1)

36-64: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Label tokens that include failed/aborted turns.

The ADR-0070 /cost contract requires token counts to be identified as including failed and aborted attempts. The current row output only says tok, which can make these counts look like completed-turn usage. Add the qualification to the row or panel header and assert it in the test.

Suggested fix

-    const tokens = `${r.inputTokens + r.outputTokens} tok`;
+    const tokens = `${r.inputTokens + r.outputTokens} tok (incl. failed/aborted turns)`;

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@apps/cli/src/chat/repl-info.ts` around lines 36 - 64, Update costNotice so
the token-count display explicitly states that counts include failed and aborted
attempts, preferably by qualifying the row’s existing “tok” label or the panel
header without altering cost or attribution behavior. Anchor the change in
costNotice’s non-legacy row formatting, and update the related test to assert
the required qualification.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@packages/db/src/session-store.test.ts`:
- Line 844: Update the SHIPPED migration directory constant in
session-store.test.ts to resolve from the test file’s directory rather than
process.cwd(), ensuring it targets packages/db/drizzle under repository-root
execution while preserving the existing migration test behavior.

---

Outside diff comments:
In `@apps/cli/src/chat/repl-info.ts`:
- Around line 36-64: Update costNotice so the token-count display explicitly
states that counts include failed and aborted attempts, preferably by qualifying
the row’s existing “tok” label or the panel header without altering cost or
attribution behavior. Anchor the change in costNotice’s non-legacy row
formatting, and update the related test to assert the required qualification.

In `@docs/decisions/0070-durable-per-model-session-cost-attribution.md`:
- Around line 60-62: Update the upsert target description in the ADR to include
the is_legacy discriminator, specifying (session_id, model, is_legacy) rather
than only (session_id, model). Preserve the additive update semantics and
explain that the discriminator keeps sentinel and custom-model records distinct.
- Line 81: Clarify the invariant for every agent_sessions row so sessions
without session_costs entries treat the empty sum as zero. Update the displayed
SUM expression to use COALESCE(SUM(...), 0), or explicitly state that an empty
sum equals zero, while preserving the existing equality to
total_cost_microcents.
- Around line 167-172: Update the `/cost` read API description to name
`SessionStore.loadSessionCostBreakdown(sessionId)` as the transactional source
for the total and per-model rows, replacing or clarifying the standalone
`loadSessionCosts(sessionId)` reference. State that the panel must obtain both
values from this single transaction snapshot rather than combining separate
reads.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 530fdd6e-13da-425b-acb9-f58daddde97a

📥 Commits

Reviewing files that changed from the base of the PR and between 38d556b and 16be31d.

📒 Files selected for processing (12)

• apps/cli/src/chat/repl-info.test.ts
• apps/cli/src/chat/repl-info.ts
• apps/cli/src/render/tui/chat-projection.test.ts
• docs/decisions/0070-durable-per-model-session-cost-attribution.md
• docs/reference/desktop/database-schema.md
• packages/db/drizzle/0009_lonely_marauders.sql
• packages/db/drizzle/0010_clumsy_jocasta.sql
• packages/db/drizzle/meta/0010_snapshot.json
• packages/db/drizzle/meta/_journal.json
• packages/db/src/schema.ts
• packages/db/src/session-store.test.ts
• packages/db/src/session-store.ts

🚧 Files skipped from review as they are similar to previous changes (6)

• packages/db/drizzle/meta/_journal.json
• packages/db/src/schema.ts
• docs/reference/desktop/database-schema.md
• packages/db/drizzle/0009_lonely_marauders.sql
• apps/cli/src/render/tui/chat-projection.test.ts
• packages/db/src/session-store.ts

[coderabbitai]

Actionable comments posted: 2

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

docs/decisions/0059-cli-mid-session-model-reseat.md (1)

105-122: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Restore the amendment numbering order.

The note enumerates items as 1, 2, 4, 3. Renumber or reorder the entries to keep the amendment sequence unambiguous.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/decisions/0059-cli-mid-session-model-reseat.md` around lines 105 - 122,
Restore sequential amendment ordering in the ADR correction note by changing the
visible item labels from 4 and 3 to the appropriate 3 and 4 order, or reorder
the entries accordingly. Keep each amendment’s content unchanged and ensure the
sequence is unambiguous.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/decisions/0059-cli-mid-session-model-reseat.md`:
- Line 112: Update the ADR-0062 link in the boundary reference to use the
identified relative target filename,
0062-context-compaction-and-cli-history-commands.md, while preserving the
existing anchor and surrounding text.
- Around line 105-115: Append an explicit supersession statement to ADR 0059
that overrides the earlier claim that file contents are not carried across model
reseat. State that `@-attached` file contents are included in the durable
user-message transcript and are carried, while tool-read file contents are not
carried because tool pairs do not cross turn boundaries; leave the existing ADR
text unchanged.

---

Outside diff comments:
In `@docs/decisions/0059-cli-mid-session-model-reseat.md`:
- Around line 105-122: Restore sequential amendment ordering in the ADR
correction note by changing the visible item labels from 4 and 3 to the
appropriate 3 and 4 order, or reorder the entries accordingly. Keep each
amendment’s content unchanged and ensure the sequence is unambiguous.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 6f004819-231d-40a3-8a86-36c3b84b4f12

📥 Commits

Reviewing files that changed from the base of the PR and between 16be31d and 8ba7737.

📒 Files selected for processing (6)

• apps/cli/src/chat/repl-info.test.ts
• apps/cli/src/chat/repl-info.ts
• apps/cli/src/commands/chat.test.ts
• apps/cli/src/home/drive-home.test.ts
• apps/cli/src/render/tui/home-app.test.tsx
• docs/decisions/0059-cli-mid-session-model-reseat.md

🚧 Files skipped from review as they are similar to previous changes (5)

• apps/cli/src/home/drive-home.test.ts
• apps/cli/src/chat/repl-info.test.ts
• apps/cli/src/chat/repl-info.ts
• apps/cli/src/commands/chat.test.ts
• apps/cli/src/render/tui/home-app.test.tsx

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@docs/roadmap/deferred-tasks.md`:
- Around line 494-496: Update the roadmap entry around the model-switch
tool-context note to comply with the English documentation guideline: either
replace the linked analysis with an English translation, or remove the analysis
link until an English version is available, while preserving the existing
deferred-task context.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 5342be6e-5637-4553-b8ed-4881b30f73aa

📥 Commits

Reviewing files that changed from the base of the PR and between 8ba7737 and a571996.

📒 Files selected for processing (3)

• docs/decisions/0059-cli-mid-session-model-reseat.md
• docs/roadmap/deferred-tasks.md
• packages/db/src/session-store.test.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• packages/db/src/session-store.test.ts

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Keep linked documentation English or omit the link.

This English roadmap entry links to an analysis explicitly marked “Turkish,” which violates the Markdown documentation language guideline. Translate the analysis before linking it, or remove the link until an English version is available.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/roadmap/deferred-tasks.md` around lines 494 - 496, Update the roadmap
entry around the model-switch tool-context note to comply with the English
documentation guideline: either replace the linked analysis with an English
translation, or remove the analysis link until an English version is available,
while preserving the existing deferred-task context.

Source: Coding guidelines

[sonarqubecloud]

Quality Gate passed

Issues
1 New issue
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud