feat(review): one-shot AI review cadence, configurable globally + per repo

apps/gittensory-ui/src/routes/docs.privacy-security.tsx

@@ -98,6 +98,7 @@ GITTENSORY_REVIEW_SCREENSHOTS="true"             # before/after visual capture f
 GITTENSORY_REVIEW_E2E_TESTS="true"               # AI-generated E2E test coverage (needs features.e2eTests too)
 
 # Global (cron / endpoint) flags, not scoped by GITTENSORY_REVIEW_REPOS.
+GITTENSORY_REVIEW_CONTINUOUS="true"              # fleet-wide default: re-review on every push (else one-shot)
 GITTENSORY_REVIEW_OPS="true"                     # read-only anomaly scan + outcome stats endpoint
 GITTENSORY_REVIEW_SELFTUNE="true"                # self-tightening tuning loop, never loosens
 GITTENSORY_REVIEW_PARITY_AUDIT="true"            # shadow-record gate-decision parity readiness

apps/gittensory-ui/src/routes/docs.tuning.tsx

@@ -138,6 +138,18 @@ function Tuning() {
           needs its own <code>features.e2eTests: true</code> override in{" "}
           <code>.gittensory.yml</code> before the feature is active for it. Per-PR.
         </li>
+        <li>
+          <code>GITTENSORY_REVIEW_CONTINUOUS</code> — fleet-wide default AI review re-trigger
+          cadence. Off by default (one-shot): AI-generated content (main review, slop advisory,
+          linked-issue satisfaction) is produced once per PR and never regenerated automatically
+          afterward — only an explicit maintainer retrigger (the PR-panel checkbox, or{" "}
+          <code>@gittensory review</code> as a maintainer) spends a fresh call. Truthy switches the
+          fleet default to continuous — every push/CI-completion/sweep re-runs AI content
+          generation. A repo's own <code>review.auto_review.cadence</code> in{" "}
+          <code>.gittensory.yml</code> always overrides this default, in either direction. Never
+          affects the deterministic gate (CI status, mergeability, static-rule blockers), which
+          always re-evaluates regardless.
+        </li>
         <li>
           <code>GITTENSORY_REVIEW_RAG</code> — retrieval-augmented context: queries the codebase
           vector index for related code and docs (callers, related modules, existing conventions)

packages/gittensory-engine/src/focus-manifest.ts

@@ -644,10 +644,25 @@ export type LabelingRule = {
   descriptionContains: string | null;
 };
 
+/** `review.auto_review.cadence` (#one-shot-review-cadence). `one_shot` = the AI-generated content (main review,
+ *  slop advisory, linked-issue satisfaction) is produced once per PR and never automatically regenerated
+ *  afterward — not on a new push, not on CI-check completion, not on a scheduled sweep tick; only an explicit
+ *  maintainer retrigger (the PR-panel checkbox or `@gittensory review` as a maintainer) spends a fresh call.
+ *  `continuous` = the traditional behavior — every trigger re-runs AI content generation, subject to each
+ *  feature's own head-SHA cache. Orthogonal to `aiReviewMode`'s enforcement-strictness axis (off/advisory/
+ *  block) — the deterministic gate (CI status, mergeability, static-rule blockers) is NEVER affected by this
+ *  and always re-evaluates on every pass regardless of cadence. */
+export const AI_REVIEW_CADENCES = ["one_shot", "continuous"] as const;
+export type AiReviewCadence = (typeof AI_REVIEW_CADENCES)[number];
+
 /** Per-repo AI review eligibility knobs under `review.auto_review`. Unset fields are byte-identical defaults. */
 export type AutoReviewConfig = {
   /** `review.auto_review.skip_drafts`: when true, draft PRs skip AI review. null (default) ⇒ drafts reviewed as today. (#2038) */
   skipDrafts: boolean | null;
+  /** `review.auto_review.cadence`: per-repo override of the AI review re-trigger cadence. null (default) ⇒
+   *  inherit the operator's fleet-wide GITTENSORY_REVIEW_CONTINUOUS default (itself "one_shot" when unset).
+   *  (#one-shot-review-cadence) */
+  cadence: AiReviewCadence | null;
   /** `review.auto_review.ignore_authors`: author-login globs whose PRs skip AI review. Empty ⇒ every author. (#2039) */
   ignoreAuthors: string[];
   /** `review.auto_review.ignore_title_keywords`: case-insensitive title substrings that skip AI review. Empty ⇒ no skip. (#2040) */
@@ -677,6 +692,7 @@ export const EMPTY_MAX_FINDINGS_CONFIG: MaxFindingsConfig = { blockers: null, ni
 
 export const EMPTY_AUTO_REVIEW_CONFIG: AutoReviewConfig = {
   skipDrafts: null,
+  cadence: null,
   ignoreAuthors: [],
   ignoreTitleKeywords: [],
   skipLabels: [],
@@ -2299,6 +2315,7 @@ function overlayMaxFindingsConfig(base: MaxFindingsConfig, override: MaxFindings
 function overlayAutoReviewConfig(base: AutoReviewConfig, override: AutoReviewConfig): AutoReviewConfig {
   return {
     skipDrafts: pickOverlayNullable(override.skipDrafts, base.skipDrafts),
+    cadence: pickOverlayNullable(override.cadence, base.cadence),
     ignoreAuthors: pickOverlayStringList(override.ignoreAuthors, base.ignoreAuthors),
     ignoreTitleKeywords: pickOverlayStringList(override.ignoreTitleKeywords, base.ignoreTitleKeywords),
     skipLabels: pickOverlayStringList(override.skipLabels, base.skipLabels),
@@ -2493,6 +2510,7 @@ function parseReviewLabelingRules(value: JsonValue | undefined, warnings: string
 function autoReviewPresent(config: AutoReviewConfig): boolean {
   return (
     config.skipDrafts !== null ||
+    config.cadence !== null ||
     config.ignoreAuthors.length > 0 ||
     config.ignoreTitleKeywords.length > 0 ||
     config.skipLabels.length > 0 ||
@@ -2514,6 +2532,7 @@ function parseAutoReviewConfig(value: JsonValue | undefined, warnings: string[])
   const record = value as Record<string, JsonValue>;
   return {
     skipDrafts: normalizeOptionalBoolean(record.skip_drafts, "review.auto_review.skip_drafts", warnings),
+    cadence: normalizeOptionalEnum(record.cadence, "review.auto_review.cadence", AI_REVIEW_CADENCES, warnings),
     ignoreAuthors: parseManifestGlobList(record.ignore_authors, "review.auto_review.ignore_authors", warnings),
     ignoreTitleKeywords: parseAutoReviewTitleKeywords(record.ignore_title_keywords, warnings),
     skipLabels: parseAutoReviewSkipLabels(record.skip_labels, warnings),
@@ -2937,6 +2956,7 @@ export function reviewConfigToJson(review: FocusManifestReviewConfig): JsonValue
   if (autoReviewPresent(review.autoReview)) {
     const autoReview: Record<string, JsonValue> = {};
     if (review.autoReview.skipDrafts !== null) autoReview.skip_drafts = review.autoReview.skipDrafts;
+    if (review.autoReview.cadence !== null) autoReview.cadence = review.autoReview.cadence;
     if (review.autoReview.ignoreAuthors.length > 0) autoReview.ignore_authors = [...review.autoReview.ignoreAuthors];
     if (review.autoReview.ignoreTitleKeywords.length > 0) autoReview.ignore_title_keywords = [...review.autoReview.ignoreTitleKeywords];
     if (review.autoReview.skipLabels.length > 0) autoReview.skip_labels = [...review.autoReview.skipLabels];

src/db/repositories.ts

@@ -4759,6 +4759,19 @@ export async function putCachedAiSlopAdvisory(
     .run();
 }
 
+/** #one-shot-review-cadence: does at least one slop-advisory row exist for this PR, regardless of head SHA?
+ *  Consulted ONLY when the resolved AI review cadence is "one_shot" — an existing row means this PR already
+ *  had its one-shot slop pass, so a later automatic trigger (push/CI-completion/sweep) must not spend another
+ *  LLM call. Existence-only (no reuse-for-display): mirrors the pre-existing commitThresholdReached silent-skip
+ *  precedent for this same feature, which also does not resurface a prior finding once paused. */
+export async function hasPublishedAiSlopAdvisory(env: Env, repoFullName: string, pullNumber: number): Promise<boolean> {
+  const row = await env.DB
+    .prepare("SELECT 1 AS present FROM ai_slop_cache WHERE repo_full_name = ? AND pull_number = ? LIMIT 1")
+    .bind(repoFullName, pullNumber)
+    .first<{ present: number }>();
+  return Boolean(row);
+}
+
 /** #linked-issue-satisfaction-cache: the stored linked-issue satisfaction result for (repo, pull, head SHA,
  *  linked issue number), or null on a miss. Mirrors getCachedAiSlopAdvisory -- every stored row is
  *  unconditionally durable (no cacheable/allowNonCacheable/maxAgeMs dimension). A nullish head SHA is always a
@@ -4814,6 +4827,25 @@ export async function putCachedLinkedIssueSatisfaction(
     .run();
 }
 
+/** #one-shot-review-cadence: does a linked-issue satisfaction row exist for this PR + linked issue number,
+ *  regardless of head SHA? Consulted ONLY when the resolved AI review cadence is "one_shot", mirroring
+ *  hasPublishedAiSlopAdvisory. Scoped ADDITIONALLY to linkedIssueNumber (not just the PR) — a PR's primary
+ *  linked issue can change between passes (see linked_issue_satisfaction_cache's own doc comment), and a
+ *  newly-linked issue has never been assessed, so it must still get its own first pass under one-shot mode
+ *  rather than being silently blocked by an unrelated issue's prior assessment. */
+export async function hasPublishedLinkedIssueSatisfaction(
+  env: Env,
+  repoFullName: string,
+  pullNumber: number,
+  linkedIssueNumber: number,
+): Promise<boolean> {
+  const row = await env.DB
+    .prepare("SELECT 1 AS present FROM linked_issue_satisfaction_cache WHERE repo_full_name = ? AND pull_number = ? AND linked_issue_number = ? LIMIT 1")
+    .bind(repoFullName, pullNumber, linkedIssueNumber)
+    .first<{ present: number }>();
+  return Boolean(row);
+}
+
 /** #4499 (grounding-file-content-cache): the stored file content for (repo, path, head SHA), or null on a
  *  miss. Unlike linked_issue_satisfaction_cache, every stored row is durable with NO input-fingerprint
  *  dimension -- file content at an immutable head SHA has exactly one correct value, so a hit is always safe

src/env.d.ts

@@ -259,6 +259,15 @@ declare global {
      *  E2E test coverage feature. Default OFF — unset/false the feature is never active for any repo regardless
      *  of a per-repo `features.e2eTests` override. */
     GITTENSORY_REVIEW_E2E_TESTS?: string;
+    /** #one-shot-review-cadence: the operator's FLEET-WIDE default for AI review re-trigger cadence, consulted
+     *  only when a repo's `.gittensory.yml review.auto_review.cadence` is unset (a per-repo value always wins
+     *  regardless of this flag — see resolveAiReviewCadence). Default OFF (unset/false) ⇒ "one_shot": the
+     *  AI-generated content (main review, slop advisory, linked-issue satisfaction) freezes after its first
+     *  pass for every repo, and only an explicit maintainer retrigger spends a fresh call. Truthy ⇒
+     *  "continuous": the traditional behavior where every push/CI-completion/sweep trigger re-runs AI content
+     *  generation, for operators who prefer that over one-shot. Never affects the deterministic gate (CI
+     *  status, mergeability, static-rule blockers), which always re-evaluates regardless of this flag. */
+    GITTENSORY_REVIEW_CONTINUOUS?: string;
     /** Convergence (reputation): when truthy, the INTERNAL-only ported submitter-reputation signal extends the
      *  AI-spend gate — a new / burst / low-reputation submitter is downgraded to a deterministic-only review
      *  (the AI neurons are skipped), and the per-(project, submitter) outcome is recorded after the gate