pierre-review 0.1.34 → 0.1.36

package/dist/api/routes/claude-review.js

@@ -7,7 +7,7 @@ import { hasUserAnthropicKey, setUserAnthropicKey, } from '../../review/local-se
 import { buildAnchorIndex, buildReview, fallbackAnchor, fetchCurrentHeadSha, fetchPrDiff, findingCommentBody, prLevelFindingBody, stripNoiseFromDiff, submitGithubComment, submitGithubIssueComment, submitGithubReview, } from '../../review/post-review.js';
 import { isNoiseFile } from '../../review/prompt.js';
 import { accountIdOf } from '../plugins/auth.js';
-const MODELS = ['claude-opus-4-8', 'claude-sonnet-4-6'];
+const MODELS = ['claude-opus-4-8', 'claude-sonnet-4-6', 'claude-haiku-4-5'];
 const VERDICTS = ['COMMENT', 'REQUEST_CHANGES', 'APPROVE'];
 const REVIEW_MODES = ['auto', 'diff_only', 'worktree'];
 const idParam = {

package/dist/config.js

@@ -29,6 +29,17 @@ function floatFromEnv(key, fallback) {
     const n = Number.parseFloat(raw);
     return Number.isFinite(n) ? n : fallback;
 }
+// The Claude Agent SDK `effort` levels (guides thinking depth + overall token
+// spend). Lower effort → fewer/cheaper thinking tokens + terser output. NOTE:
+// `effort` is rejected by Haiku 4.5 — only models that accept it get it (see
+// review/agent.ts EFFORT_CAPABLE_MODELS).
+const REVIEW_EFFORTS = ['low', 'medium', 'high', 'xhigh', 'max'];
+function effortFromEnv(key, fallback) {
+    const raw = process.env[key];
+    return REVIEW_EFFORTS.includes(raw ?? '')
+        ? raw
+        : fallback;
+}
 // ---- Deployment mode (the master switch) ----
 // `local` (default): SQLite via better-sqlite3, `gh auth token` auth, one
 // implicit account, no landing page — the unchanged zero-config experience.
@@ -122,10 +133,41 @@ export const config = {
     // Per-run caps (cost/disk/time runaway guards). The diff is inlined in full, so
     // reviews need far fewer turns than the old default; 30 is still generous.
     reviewMaxTurns: intFromEnv('REVIEW_MAX_TURNS', 30),
-    reviewBudgetUsd: floatFromEnv('REVIEW_BUDGET_USD', 1.0),
+    // Hard USD ceiling per run. When the SDK trips this it returns an
+    // `error_max_budget_usd` result BEFORE the agent calls submit_review — so a run
+    // that hits the cap is recorded FAILED and still bills (you pay for no review).
+    // The cap must therefore sit ABOVE the cost of a normal completed review, not at
+    // it; `reviewEffort` below is the real cost lever (a large diff at default-high
+    // effort is what blew the old $1 cap). Lower REVIEW_BUDGET_USD only if you'd
+    // rather a borderline review fail than complete.
+    reviewBudgetUsd: floatFromEnv('REVIEW_BUDGET_USD', 1.5),
     // Turn cap for a diff-only run. These are TOOL-LESS (only submit_review), so they
     // should finish in ~2 turns; a tight cap is a cheap runaway guard.
     reviewDiffOnlyMaxTurns: intFromEnv('REVIEW_DIFF_ONLY_MAX_TURNS', 6),
+    // Haiku reaches a conclusion in MORE steps than Sonnet/Opus (smaller model, more
+    // tool round-trips), so it routinely tripped the turn cap mid-review and failed.
+    // Give it proportionally more turns in both modes. Its low per-token price means
+    // the extra turns are cheap, and maxBudgetUsd is still the real spend guard.
+    reviewHaikuTurnMultiplier: floatFromEnv('REVIEW_HAIKU_TURN_MULTIPLIER', 2),
+    // ---- Diff-size cap ----
+    // A very large inlined diff is the dominant cost on a big PR (it's the cached
+    // prefix re-read every turn). The diff shown IN THE PROMPT is truncated at a
+    // whole-file boundary to this many characters; routing + line anchoring still use
+    // the FULL diff, and the changed-file LIST stays complete (so a worktree run can
+    // Read the omitted files). ON by default — it proved its worth on a large PR that
+    // failed without it. ~60k chars ≈ ~15k tokens, so only outlier PRs are truncated
+    // and a normal review is unaffected; each run records diffCapped + the full diff
+    // size, so you can still A/B by setting REVIEW_DIFF_CAP_ENABLED=false for a
+    // baseline run and comparing the recorded cost.
+    reviewDiffCapEnabled: process.env.REVIEW_DIFF_CAP_ENABLED !== 'false',
+    reviewDiffCapChars: intFromEnv('REVIEW_DIFF_CAP_CHARS', 60000),
+    // Agent `effort` per mode — the dominant cost knob (unset ⇒ the SDK default
+    // `high`, which over-thinks bounded reviews). A diff-only run just hunts bugs in a
+    // small inlined diff, so `low` is plenty; a worktree run reasons across files, so
+    // `medium` keeps that while trimming the over-exploration that ran up the bill.
+    // Applied only to effort-capable models (Sonnet/Opus); Haiku ignores it.
+    reviewEffort: effortFromEnv('REVIEW_EFFORT', 'medium'),
+    reviewDiffOnlyEffort: effortFromEnv('REVIEW_DIFF_ONLY_EFFORT', 'low'),
     // At most one review per PR; this caps concurrent reviews across all PRs. Default
     // 4 so the user can bulk-review (extras queue, see review-manager). Raising this
     // also DISABLES the pasted-key override (which mutates process.env and is only

package/dist/db/migrations/0020_claude_review_cost_telemetry.sql

@@ -0,0 +1,11 @@
+-- Claude Review cost telemetry (additive). Records the cache-token split (a
+-- multi-turn run's input is mostly cache reads — the dominant cost the plain
+-- input_tokens column hid) plus the full noise-stripped diff size and whether the
+-- feature-flagged diff-size cap truncated the prompt (so capped vs uncapped runs can
+-- be cost-compared). All nullable; existing rows stay NULL. SQLite-only: Claude
+-- Review is force-disabled in cloud, so the Postgres claude_reviews table is never
+-- populated (its baseline is regenerated separately via db:generate:pg).
+ALTER TABLE `claude_reviews` ADD `cache_read_tokens` integer;--> statement-breakpoint
+ALTER TABLE `claude_reviews` ADD `cache_creation_tokens` integer;--> statement-breakpoint
+ALTER TABLE `claude_reviews` ADD `diff_bytes` integer;--> statement-breakpoint
+ALTER TABLE `claude_reviews` ADD `diff_capped` integer;

package/dist/db/migrations/meta/_journal.json

@@ -141,6 +141,13 @@
       "when": 1780800000010,
       "tag": "0019_finding_file_in_diff",
       "breakpoints": true
+    },
+    {
+      "idx": 20,
+      "version": "6",
+      "when": 1780800000011,
+      "tag": "0020_claude_review_cost_telemetry",
+      "breakpoints": true
     }
   ]
 }

package/dist/db/migrations-pg/0009_fuzzy_jubilee.sql

@@ -0,0 +1,4 @@
+ALTER TABLE "claude_reviews" ADD COLUMN "cache_read_tokens" integer;--> statement-breakpoint
+ALTER TABLE "claude_reviews" ADD COLUMN "cache_creation_tokens" integer;--> statement-breakpoint
+ALTER TABLE "claude_reviews" ADD COLUMN "diff_bytes" integer;--> statement-breakpoint
+ALTER TABLE "claude_reviews" ADD COLUMN "diff_capped" boolean;