gsd-pi 2.76.0 → 2.77.0-dev.1594b263e

package/dist/resources/extensions/gsd/memory-backfill.js

@@ -0,0 +1,126 @@
+// GSD2 — Decisions -> memories backfill (ADR-013 step 5)
+//
+// Idempotent one-shot migration that copies every active decisions row into
+// the memories table with category="architecture" and a structured_fields
+// payload preserving the original gsd_save_decision schema (when_context,
+// scope, decision, choice, rationale, made_by, revisable, sourceDecisionId).
+//
+// The backfill exists so the cutover in ADR-013 step 6 can drop the
+// decisions table without losing schema fidelity. Idempotency is enforced
+// by tagging each backfilled memory with structured_fields.sourceDecisionId
+// and skipping any decision whose ID already appears in the memories table.
+//
+// Triggered opportunistically by buildBeforeAgentStartResult so the cost
+// only ever fires once per project. Costs O(N) inserts on first run where
+// N is the active-decisions count; subsequent runs are an O(N) lookup that
+// finds existing markers and exits.
+import { isDbAvailable, _getAdapter } from "./gsd-db.js";
+import { createMemory } from "./memory-store.js";
+import { logWarning } from "./workflow-logger.js";
+/**
+ * Backfill active decisions rows into the memories table.
+ *
+ * - Idempotent (per-row): every row written carries
+ *   `structured_fields.sourceDecisionId = "<decisionId>"`. Each candidate
+ *   decision is checked individually; only decisions whose id is already
+ *   present in the memory store are skipped. A user-authored memory with
+ *   their own `sourceDecisionId` does NOT abort the backfill.
+ * - Best-effort: never throws. Logs and returns 0 on failure so a broken
+ *   backfill cannot block agent startup.
+ * - Active-only: skips rows where `superseded_by IS NOT NULL`. Superseded
+ *   decisions are historical record; the memory store is for active
+ *   knowledge.
+ *
+ * Returns the number of memories written (0 when already backfilled or
+ * when the DB has no decisions). Callers can log the result or surface it
+ * to the user.
+ */
+export function backfillDecisionsToMemories() {
+    if (!isDbAvailable())
+        return 0;
+    const adapter = _getAdapter();
+    if (!adapter)
+        return 0;
+    try {
+        const decisions = adapter
+            .prepare("SELECT id, when_context, scope, decision, choice, rationale, made_by, revisable, superseded_by FROM decisions WHERE superseded_by IS NULL")
+            .all();
+        if (decisions.length === 0)
+            return 0;
+        // Per-row idempotency: each memory backfilled from a decision carries
+        // structured_fields.sourceDecisionId="<decisionId>". Skipping is decided
+        // per row by matching that exact id, NOT by a global sentinel — a global
+        // sentinel would silently abort the backfill if a user manually called
+        // capture_thought with their own structuredFields.sourceDecisionId.
+        const checkExisting = adapter.prepare("SELECT 1 FROM memories WHERE structured_fields LIKE :pattern LIMIT 1");
+        let written = 0;
+        for (const raw of decisions) {
+            const row = {
+                id: String(raw["id"] ?? ""),
+                when_context: String(raw["when_context"] ?? ""),
+                scope: String(raw["scope"] ?? ""),
+                decision: String(raw["decision"] ?? ""),
+                choice: String(raw["choice"] ?? ""),
+                rationale: String(raw["rationale"] ?? ""),
+                made_by: String(raw["made_by"] ?? "agent"),
+                revisable: String(raw["revisable"] ?? ""),
+                superseded_by: raw["superseded_by"] == null ? null : String(raw["superseded_by"]),
+            };
+            if (!row.id)
+                continue;
+            // Pattern is anchored to the JSON-stringified shape and the exact
+            // decision id to avoid prefix collisions (e.g. "D1" vs "D10").
+            if (checkExisting.get({ ":pattern": `%"sourceDecisionId":"${row.id}"%` }))
+                continue;
+            const content = synthesizeContent(row);
+            const id = createMemory({
+                category: "architecture",
+                content,
+                scope: row.scope || "project",
+                confidence: 0.85,
+                structuredFields: {
+                    sourceDecisionId: row.id,
+                    when_context: row.when_context,
+                    scope: row.scope,
+                    decision: row.decision,
+                    choice: row.choice,
+                    rationale: row.rationale,
+                    made_by: row.made_by,
+                    revisable: row.revisable,
+                },
+            });
+            if (id)
+                written += 1;
+        }
+        return written;
+    }
+    catch (e) {
+        logWarning("memory-backfill", `decisions->memories backfill failed: ${e.message}`);
+        return 0;
+    }
+}
+/**
+ * Combine the decision's structured fields into a 1-3 sentence content
+ * string suitable for keyword retrieval and human review.
+ *
+ * Format: "<decision> Chose: <choice>. Rationale: <rationale>."
+ * Truncates each field to keep the synthesized line under ~600 chars so
+ * memory_query rendering stays readable.
+ */
+function synthesizeContent(row) {
+    const trim = (value, max) => {
+        const cleaned = value.replace(/\s+/g, " ").trim();
+        return cleaned.length > max ? cleaned.slice(0, max - 1) + "\u2026" : cleaned;
+    };
+    const parts = [];
+    const decision = trim(row.decision, 240);
+    const choice = trim(row.choice, 200);
+    const rationale = trim(row.rationale, 200);
+    if (decision)
+        parts.push(decision);
+    if (choice)
+        parts.push(`Chose: ${choice}.`);
+    if (rationale)
+        parts.push(`Rationale: ${rationale}.`);
+    return parts.join(" ");
+}

package/dist/resources/extensions/gsd/memory-extractor.js

@@ -62,7 +62,9 @@ export function buildMemoryLLMCall(ctx) {
         // which returns undefined for OAuth users (Claude Max / Claude Pro).
         // See: https://github.com/gsd-build/gsd-2/issues/2959
         const resolvedKeyPromise = ctx.modelRegistry.getApiKey(selectedModel).catch(() => undefined);
-        return async (system, user) => {
+        // Expose on the returned fn so tests can await resolution deterministically
+        // (avoids arbitrary setTimeout polling for an internal microtask).
+        const llmCall = async (system, user) => {
             const { completeSimple } = await import('@gsd/pi-ai');
             const resolvedApiKey = await resolvedKeyPromise;
             const result = await completeSimple(selectedModel, {
@@ -79,6 +81,10 @@ export function buildMemoryLLMCall(ctx) {
                 .map(c => c.text);
             return textParts.join('');
         };
+        // Attach the in-flight API-key resolution so tests (and callers) can
+        // `await llmCall.apiKeyReady` rather than relying on setTimeout polling.
+        llmCall.apiKeyReady = resolvedKeyPromise;
+        return llmCall;
     }
     catch {
         return null;

package/dist/resources/extensions/gsd/memory-store.js

@@ -29,8 +29,22 @@ function rowToMemory(row) {
         hit_count: row['hit_count'],
         scope: row['scope'] ?? 'project',
         tags: parseTags(row['tags']),
+        structured_fields: parseStructuredFields(row['structured_fields']),
     };
 }
+function parseStructuredFields(raw) {
+    if (typeof raw !== 'string' || raw.length === 0)
+        return null;
+    try {
+        const parsed = JSON.parse(raw);
+        return parsed && typeof parsed === 'object' && !Array.isArray(parsed)
+            ? parsed
+            : null;
+    }
+    catch {
+        return null;
+    }
+}
 function parseTags(raw) {
     if (typeof raw !== 'string' || raw.length === 0)
         return [];
@@ -376,6 +390,7 @@ export function createMemory(fields) {
             updatedAt: now,
             scope: fields.scope ?? 'project',
             tags: fields.tags ?? [],
+            structuredFields: fields.structuredFields ?? null,
         });
         // Derive the real ID from the assigned seq (SELECT is still fine via adapter)
         const row = adapter.prepare('SELECT seq FROM memories WHERE id = :id').get({ ':id': placeholder });
@@ -553,6 +568,10 @@ export function applyMemoryActions(actions, unitType, unitId) {
                             source_unit_id: unitId,
                             scope: action.scope,
                             tags: action.tags,
+                            // ADR-013: forward structured payload through the action layer so
+                            // bulk applyMemoryActions callers (extraction, ingestion) don't
+                            // silently drop it.
+                            structuredFields: action.structuredFields ?? null,
                         });
                         break;
                     case 'UPDATE':

package/dist/resources/extensions/gsd/milestone-actions.js

@@ -18,6 +18,18 @@ import { loadQueueOrder, saveQueueOrder } from "./queue-order.js";
 import { deleteMilestone, getMilestone, isDbAvailable, updateMilestoneStatus } from "./gsd-db.js";
 import { removeWorktree } from "./worktree-manager.js";
 import { logWarning } from "./workflow-logger.js";
+import { isAutoActive } from "./auto.js";
+/**
+ * Writer-side assert for mutations that race with auto-mode's squash merge (#4704).
+ * Auto-mode is confirmed not to call parkMilestone/discardMilestone/unparkMilestone
+ * internally — these throws only surface invariant violations from new or forgotten
+ * call sites, which is the correct failure mode to catch loudly.
+ */
+function assertNotAutoActive(action) {
+    if (isAutoActive()) {
+        throw new Error(`${action} cannot run while auto-mode is active. Stop auto-mode first with /gsd stop.`);
+    }
+}
 // ─── Park ──────────────────────────────────────────────────────────────────
 /**
  * Park a milestone — creates a PARKED.md marker file with reason and timestamp.
@@ -25,6 +37,7 @@ import { logWarning } from "./workflow-logger.js";
  * Returns true if successfully parked, false if milestone not found or already parked.
  */
 export function parkMilestone(basePath, milestoneId, reason) {
+    assertNotAutoActive("park milestone");
     const mDir = resolveMilestonePath(basePath, milestoneId);
     if (!mDir || !existsSync(mDir))
         return false;
@@ -65,6 +78,7 @@ export function parkMilestone(basePath, milestoneId, reason) {
  * Returns true if successfully unparked, false if milestone not found or not parked.
  */
 export function unparkMilestone(basePath, milestoneId) {
+    assertNotAutoActive("unpark milestone");
     const mDir = resolveMilestonePath(basePath, milestoneId);
     if (!mDir || !existsSync(mDir))
         return false;
@@ -97,6 +111,7 @@ export function unparkMilestone(basePath, milestoneId) {
  * Returns true if successfully discarded, false if milestone not found.
  */
 export function discardMilestone(basePath, milestoneId) {
+    assertNotAutoActive("discard milestone");
     const mDir = resolveMilestonePath(basePath, milestoneId);
     if (!mDir || !existsSync(mDir))
         return false;

package/dist/resources/extensions/gsd/milestone-scope-classifier.js

@@ -0,0 +1,299 @@
+// GSD-2 — Milestone scope classifier (#4781 / ADR-003 companion).
+//
+// Pure heuristics over milestone planning fields. Produces a PipelineVariant
+// that downstream dispatch logic can use to shape the auto-mode sequence.
+// No LLM calls, no file I/O, sub-millisecond.
+//
+// Distinct from `complexity-classifier.ts`, which decides *model tier*
+// (light/standard/heavy) for an individual unit. This module decides
+// *pipeline topology* for an entire milestone at plan-milestone time.
+//
+// This file ships the classifier in isolation. Dispatch-side wiring
+// lands in follow-up PRs so the classification contract can be reviewed
+// and tested before any behavior change reaches users.
+// ─── Keyword sets ─────────────────────────────────────────────────────────
+/**
+ * Override keywords that force `standard` (at minimum) regardless of
+ * apparent triviality. Presence of any of these signals work that is
+ * either security-sensitive, irreversible, or requires runtime verification
+ * a "trivial" pipeline would skip.
+ *
+ * Matched as case-insensitive word-boundary substrings. Conservative — err
+ * on the side of including a keyword; over-classifying to `standard` costs
+ * units, under-classifying could ship broken auth/security/migration work.
+ */
+const OVERRIDE_KEYWORDS = [
+    // Security-sensitive
+    "security", "auth", "authn", "authz", "authentication", "authorization",
+    "credential", "secret", "password", "token", "oauth", "encrypt", "decrypt",
+    "vulnerability", "exploit", "permission", "rbac", "acl",
+    // Data-migration / irreversible
+    "migration", "migrate", "schema change", "data migration",
+    "backfill", "drop column", "drop table",
+    // Compliance / regulatory
+    "compliance", "gdpr", "hipaa", "soc2", "pci",
+    // Infra / deploy — runtime verification needed
+    "deploy", "rollout", "canary", "production database",
+];
+/**
+ * Keywords that contribute to `complex` classification on their own.
+ * Different from OVERRIDE_KEYWORDS in that a single match bumps to
+ * complex, not just to standard.
+ */
+const COMPLEX_KEYWORDS = [
+    "multi-service", "distributed", "consensus", "saga", "eventual consistency",
+    "breaking change", "api contract change", "schema redesign",
+    "architect", "architecture", "refactor core",
+];
+/**
+ * Trivial-signal keywords: presence strongly suggests a simple, contained
+ * deliverable. Only effective when combined with low file count / no tests
+ * / no override keywords.
+ */
+const TRIVIAL_KEYWORDS = [
+    "single file", "one file", "static html", "static page",
+    "one-page", "landing page", "readme", "docs only", "typo", "rename",
+    "spelling", "comment", "changelog",
+    // Browser-only / no-build deliverable shapes (b23 forensic case).
+    "pure html", "browser-based", "no build step", "no build tooling",
+    "localstorage", "client-only", "no backend", "no server", "no backend.",
+];
+// ─── Heuristics ───────────────────────────────────────────────────────────
+/**
+ * Estimate how many distinct files the milestone will touch, based on
+ * explicit mentions in the input text. Returns `null` when no hint is
+ * discoverable — callers should treat that as "unknown, no signal."
+ */
+function extractFileCountHint(text) {
+    // Explicit phrasing: "a single file", "two files", "3 files"
+    const singleFileMatch = /\b(a|one|single)\s+(file|page)\b/i.test(text);
+    if (singleFileMatch)
+        return 1;
+    const digitMatch = text.match(/\b(\d+)\s+files?\b/i);
+    if (digitMatch) {
+        const n = parseInt(digitMatch[1], 10);
+        if (!Number.isNaN(n))
+            return n;
+    }
+    const wordMatch = text.match(/\b(two|three|four|five|six|seven|eight|nine|ten)\s+files?\b/i);
+    if (wordMatch) {
+        const wordMap = {
+            two: 2, three: 3, four: 4, five: 5,
+            six: 6, seven: 7, eight: 8, nine: 9, ten: 10,
+        };
+        return wordMap[wordMatch[1].toLowerCase()] ?? null;
+    }
+    return null;
+}
+function escapeRegExp(value) {
+    return value.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+}
+function containsAnyKeyword(haystack, keywords) {
+    const lower = haystack.toLowerCase();
+    const hits = [];
+    for (const kw of keywords) {
+        // Word-boundary match to prevent substring collisions (e.g. "auth"
+        // must not match "author", "api" must not match "capital"). Phrases
+        // containing non-word characters (hyphens, slashes) still work because
+        // `\b` sits at the word-char / non-word-char transition, so
+        // `\bbrowser-based\b` matches "browser-based" bounded by whitespace
+        // or punctuation on either side.
+        const pattern = new RegExp(String.raw `\b${escapeRegExp(kw)}\b`, "i");
+        if (pattern.test(lower))
+            hits.push(kw);
+    }
+    return hits;
+}
+/**
+ * True when `term` appears in the text without an immediately preceding
+ * negator (no / without / not / zero / skip) in the same clause. Used to
+ * keep phrases like "no backend" or "no tests" from flipping a trivial-
+ * class milestone to standard. Best-effort; imperfect English parsing,
+ * biased toward false negatives (if unsure, treats term as present —
+ * which routes to standard, the safe pipeline).
+ */
+function mentionsWithoutNegation(text, term) {
+    const lower = text.toLowerCase();
+    const termPattern = new RegExp(String.raw `\b${term}\b`, "gi");
+    const matches = Array.from(lower.matchAll(termPattern));
+    for (const m of matches) {
+        const start = m.index ?? 0;
+        const windowStart = Math.max(0, start - 30);
+        const window = lower.slice(windowStart, start);
+        // Negator anywhere in the 30-char lookback window counts as negation —
+        // covers "no backend", "without a server", "not using api", "zero
+        // dependencies on an api". If a sentence break intervenes between the
+        // negator and the term, treat as a different clause (positive mention).
+        const hasNegator = /(^|[^a-z0-9])(no|without|not|zero|skip(s|ping)?|drops?)\b/i.test(window);
+        const hasSentenceBreak = /[.;!?]/.test(window);
+        if (hasNegator && !hasSentenceBreak)
+            continue;
+        return true;
+    }
+    return false;
+}
+function mentionsTests(haystack) {
+    return mentionsWithoutNegation(haystack, "test")
+        || mentionsWithoutNegation(haystack, "tests")
+        || mentionsWithoutNegation(haystack, "testing")
+        || mentionsWithoutNegation(haystack, "spec")
+        || mentionsWithoutNegation(haystack, "unit test")
+        || mentionsWithoutNegation(haystack, "integration test");
+}
+function mentionsBackend(haystack) {
+    return mentionsWithoutNegation(haystack, "api")
+        || mentionsWithoutNegation(haystack, "backend")
+        || mentionsWithoutNegation(haystack, "server")
+        || mentionsWithoutNegation(haystack, "database")
+        || mentionsWithoutNegation(haystack, "endpoint");
+}
+// ─── DB adapter ───────────────────────────────────────────────────────────
+/**
+ * Shape adapter: convert a milestone DB row into the classifier input
+ * object. Keeps `milestone-scope-classifier.ts` free of DB types at the
+ * module boundary — callers can either use this helper or hand-build the
+ * input themselves (e.g. from plan-milestone tool params before insert).
+ */
+export function milestoneRowToScopeInput(row) {
+    return {
+        title: row.title,
+        vision: row.vision,
+        successCriteria: row.success_criteria,
+        keyRisks: row.key_risks,
+        definitionOfDone: row.definition_of_done,
+        requirementCoverage: row.requirement_coverage,
+        verificationContract: row.verification_contract,
+        verificationIntegration: row.verification_integration,
+        verificationOperational: row.verification_operational,
+        verificationUat: row.verification_uat,
+    };
+}
+/**
+ * Compute the pipeline variant for a milestone by reading its planning
+ * fields from the DB and running the classifier. Returns `null` when
+ * classification is unavailable (DB closed, milestone missing, unexpected
+ * error) — callers MUST treat null as "run the full pipeline" so a
+ * classification failure never silently downshifts dispatch.
+ */
+export async function getMilestonePipelineVariant(mid) {
+    try {
+        const { isDbAvailable, getMilestone } = await import("./gsd-db.js");
+        if (!isDbAvailable())
+            return null;
+        const row = getMilestone(mid);
+        if (!row)
+            return null;
+        return classifyMilestoneScope(milestoneRowToScopeInput(row)).variant;
+    }
+    catch {
+        return null;
+    }
+}
+// ─── Public API ───────────────────────────────────────────────────────────
+/**
+ * Classify a milestone's pipeline variant based on its planning inputs.
+ *
+ * Precedence (matches implementation order — complex-first so that
+ * security-sensitive architecture refactors correctly route to complex
+ * rather than standard; the override hit is still recorded in
+ * `signals.triggeredOverride` for telemetry):
+ *  1. Complex-signal keyword OR ≥ 8 file hint OR architecture/refactor-core
+ *     language → `complex`.
+ *  2. Override keyword → `standard` (at minimum). Prevents trivial
+ *     misclassification of security / auth / migration work.
+ *  3. Trivial-signal keyword AND ≤ 2 file hint AND no tests mentioned AND
+ *     no backend mentioned → `trivial`.
+ *  4. Otherwise → `standard`.
+ *
+ * Ambiguity → `standard` (today's default). Safe to run the full pipeline.
+ */
+export function classifyMilestoneScope(input) {
+    const haystack = [
+        input.title ?? "",
+        input.vision ?? "",
+        (input.successCriteria ?? []).join("\n"),
+        (input.keyRisks ?? []).map(r => `${r.risk ?? ""} ${r.whyItMatters ?? ""}`).join("\n"),
+        (input.definitionOfDone ?? []).join("\n"),
+        input.requirementCoverage ?? "",
+        input.verificationContract ?? "",
+        input.verificationIntegration ?? "",
+        input.verificationOperational ?? "",
+        input.verificationUat ?? "",
+    ].join("\n");
+    const overrideHits = containsAnyKeyword(haystack, OVERRIDE_KEYWORDS);
+    const complexHits = containsAnyKeyword(haystack, COMPLEX_KEYWORDS);
+    const trivialHits = containsAnyKeyword(haystack, TRIVIAL_KEYWORDS);
+    const fileCountHint = extractFileCountHint(haystack);
+    const hasTests = mentionsTests(haystack);
+    const hasBackend = mentionsBackend(haystack);
+    const reasons = [];
+    // Rule 2: complex-class signals. Evaluated before override because a
+    // complex + override input should land in complex, not standard.
+    if (complexHits.length > 0) {
+        reasons.push(`complex keywords: ${complexHits.slice(0, 3).join(", ")}`);
+    }
+    if (fileCountHint !== null && fileCountHint >= 8) {
+        reasons.push(`file count hint: ${fileCountHint}`);
+    }
+    const isComplex = complexHits.length > 0 || (fileCountHint !== null && fileCountHint >= 8);
+    if (isComplex) {
+        return {
+            variant: "complex",
+            reasons,
+            signals: {
+                triggeredOverride: overrideHits.length > 0,
+                complexCount: complexHits.length,
+                trivialCount: trivialHits.length,
+                fileCountHint,
+            },
+        };
+    }
+    // Rule 1: override keywords force standard.
+    if (overrideHits.length > 0) {
+        return {
+            variant: "standard",
+            reasons: [`override keywords: ${overrideHits.slice(0, 3).join(", ")}`],
+            signals: {
+                triggeredOverride: true,
+                complexCount: complexHits.length,
+                trivialCount: trivialHits.length,
+                fileCountHint,
+            },
+        };
+    }
+    // Rule 3: trivial signals — require ALL of: trivial-keyword, low file
+    // hint (or nothing suggesting high count), no test mention, no backend
+    // mention.
+    const fileCountOk = fileCountHint === null || fileCountHint <= 2;
+    const trivial = trivialHits.length > 0 &&
+        fileCountOk &&
+        !hasTests &&
+        !hasBackend;
+    if (trivial) {
+        reasons.push(`trivial keywords: ${trivialHits.slice(0, 3).join(", ")}`);
+        if (fileCountHint !== null)
+            reasons.push(`file count hint: ${fileCountHint}`);
+        reasons.push("no tests mentioned", "no backend mentioned");
+        return {
+            variant: "trivial",
+            reasons,
+            signals: {
+                triggeredOverride: false,
+                complexCount: complexHits.length,
+                trivialCount: trivialHits.length,
+                fileCountHint,
+            },
+        };
+    }
+    // Rule 4: fallback.
+    return {
+        variant: "standard",
+        reasons: reasons.length > 0 ? reasons : ["no strong signals — default"],
+        signals: {
+            triggeredOverride: overrideHits.length > 0,
+            complexCount: complexHits.length,
+            trivialCount: trivialHits.length,
+            fileCountHint,
+        },
+    };
+}

package/dist/resources/extensions/gsd/milestone-summary-classifier.js

@@ -0,0 +1,37 @@
+/**
+ * Shared milestone SUMMARY classifier.
+ *
+ * SUMMARY presence alone is not enough to prove milestone completion: recovery
+ * and blocker paths also write SUMMARY files. Keep this leaf module free of
+ * state/auto imports so state derivation, dispatch guards, and recovery can
+ * share one definition without cycles.
+ */
+import { splitFrontmatter, parseFrontmatterMap } from "../shared/frontmatter.js";
+import { isClosedStatus } from "./status-guards.js";
+export function classifyMilestoneSummaryContent(content) {
+    const [fmLines] = splitFrontmatter(content);
+    const fm = fmLines ? parseFrontmatterMap(fmLines) : null;
+    const rawStatus = typeof fm?.status === "string" ? fm.status.trim().toLowerCase() : "";
+    if (rawStatus) {
+        if (isClosedStatus(rawStatus))
+            return "success";
+        if (["active", "pending", "blocked", "failed", "failure", "incomplete"].includes(rawStatus)) {
+            return "failure";
+        }
+    }
+    const failureSignal = /(?:^|\n)\s*#\s*BLOCKER\b/i.test(content)
+        || /auto-mode recovery failed/i.test(content)
+        || /verification\s+failed/i.test(content)
+        || /(?:^|\n)\s*(?:status|verdict|outcome|result)\s*[:=-]\s*not complete\b/i.test(content);
+    if (failureSignal)
+        return "failure";
+    return "unknown";
+}
+/**
+ * Legacy-compatible terminal check for state derivation.
+ * Unknown summaries remain terminal to preserve old handwritten SUMMARY files;
+ * explicit failure summaries do not.
+ */
+export function isTerminalMilestoneSummaryContent(content) {
+    return classifyMilestoneSummaryContent(content) !== "failure";
+}

package/dist/resources/extensions/gsd/model-cost-table.js

@@ -43,6 +43,9 @@ export const BUNDLED_COST_TABLE = [
     { id: "gpt-5.3-codex-spark", inputPer1k: 0.0003, outputPer1k: 0.0012, updatedAt: "2026-03-29" },
     { id: "gpt-5.4", inputPer1k: 0.005, outputPer1k: 0.02, updatedAt: "2026-03-29" },
     { id: "gpt-5.4-mini", inputPer1k: 0.00075, outputPer1k: 0.0045, updatedAt: "2026-04-18" },
+    // GPT-5.5 API list price, also used for live Codex OAuth routing.
+    // Source: https://openai.com/api/pricing/
+    { id: "gpt-5.5", inputPer1k: 0.005, outputPer1k: 0.03, updatedAt: "2026-04-23" },
     // Google
     { id: "gemini-2.0-flash", inputPer1k: 0.0001, outputPer1k: 0.0004, updatedAt: "2025-03-15" },
     { id: "gemini-flash-2.0", inputPer1k: 0.0001, outputPer1k: 0.0004, updatedAt: "2025-03-15" },