gsd-pi 2.77.0 → 2.78.0-dev.aeeb2ca00

package/src/resources/extensions/gsd/milestone-summary-classifier.ts

@@ -0,0 +1,42 @@
+/**
+ * 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 type MilestoneSummaryOutcome = "success" | "failure" | "unknown";
+
+export function classifyMilestoneSummaryContent(content: string): MilestoneSummaryOutcome {
+  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: string): boolean {
+  return classifyMilestoneSummaryContent(content) !== "failure";
+}

package/src/resources/extensions/gsd/model-cost-table.ts

@@ -57,6 +57,9 @@ export const BUNDLED_COST_TABLE: ModelCostEntry[] = [
   { 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" },

package/src/resources/extensions/gsd/model-router.ts

@@ -109,6 +109,7 @@ export const MODEL_CAPABILITY_TIER: Record<string, ComplexityTier> = {
   "gpt-5.2-codex": "heavy",
   "gpt-5.3-codex": "heavy",
   "gpt-5.4": "heavy",
+  "gpt-5.5": "heavy",
   "o1": "heavy",
   "o3": "heavy",
   "o4-mini": "heavy",
@@ -144,6 +145,7 @@ const MODEL_COST_PER_1K_INPUT: Record<string, number> = {
   "gpt-5.3-codex": 0.005,
   "gpt-5.3-codex-spark": 0.0003,
   "gpt-5.4": 0.005,
+  "gpt-5.5": 0.005,
   "o4-mini": 0.005,
   "o4-mini-deep-research": 0.005,
   "gemini-2.0-flash": 0.0001,
@@ -187,6 +189,10 @@ export const MODEL_CAPABILITY_PROFILES: Record<string, ModelCapabilities> = {
   "gpt-5.3-codex":                { coding: 94, debugging: 91, research: 74, reasoning: 89, speed: 50, longContext: 80, instruction: 89 },
   "gpt-5.3-codex-spark":          { coding: 68, debugging: 58, research: 42, reasoning: 52, speed: 90, longContext: 50, instruction: 74 },
   "gpt-5.4":                      { coding: 95, debugging: 92, research: 88, reasoning: 94, speed: 42, longContext: 88, instruction: 92 },
+  // GPT-5.5 scores are relative to the existing gpt-5.4 profile and backed by
+  // OpenAI's 2026-04-23 published eval deltas across coding, tool use, and long context.
+  // Source: https://openai.com/index/introducing-gpt-5-5/
+  "gpt-5.5":                      { coding: 96, debugging: 93, research: 89, reasoning: 95, speed: 42, longContext: 90, instruction: 93 },
 
   // ── OpenAI o-series (reasoning-first) ──────────────────────────────────────
   "o1":                           { coding: 78, debugging: 82, research: 78, reasoning: 90, speed: 20, longContext: 65, instruction: 82 },

package/src/resources/extensions/gsd/native-git-bridge.ts

@@ -894,11 +894,37 @@ export function nativeResetPaths(basePath: string, paths: string[]): void {
   }
 }
 
+/**
+ * Read `commit.gpgsign` from the repo config. Returns true only if the value
+ * is the literal string "true". Any other state (unset, false, error) → false.
+ *
+ * Used by nativeCommit to route signing-required commits through the git CLI,
+ * because the libgit2 native path does not invoke configured signers.
+ * (Issue #4980 CRIT-2)
+ */
+function shouldSignCommits(basePath: string): boolean {
+  try {
+    const result = execFileSync("git", ["config", "--get", "commit.gpgsign"], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+      encoding: "utf-8",
+      env: GIT_NO_PROMPT_ENV,
+    }).trim();
+    return result === "true";
+  } catch {
+    return false;
+  }
+}
+
 /**
  * Create a commit from the current index.
  * Returns the commit SHA on success, or null if nothing to commit.
  * Native: libgit2 commit create.
- * Fallback: `git commit --no-verify -F -`.
+ * Fallback: `git commit -F -` (runs hooks; honors commit.gpgsign).
+ *
+ * The fallback intentionally does NOT use --no-verify — user pre-commit /
+ * commit-msg / prepare-commit-msg hooks must fire on every GSD-automated
+ * commit. (Issue #4980 CRIT-1)
  */
 export function nativeCommit(
   basePath: string,
@@ -906,7 +932,10 @@ export function nativeCommit(
   options?: { allowEmpty?: boolean; input?: string },
 ): string | null {
   const native = loadNative();
-  if (native) {
+  // libgit2's commit-create does not invoke configured GPG/SSH signers. When
+  // commit.gpgsign=true, route through the git CLI fallback so signing
+  // happens. (Issue #4980 CRIT-2)
+  if (native && !shouldSignCommits(basePath)) {
     try {
       return native.gitCommit(basePath, message, options?.allowEmpty);
     } catch (e) {
@@ -916,9 +945,10 @@ export function nativeCommit(
     }
   }
 
-  // Fallback: use git commit with stdin pipe for safe multi-line messages
+  // Fallback / signed-commit path: use git CLI with stdin pipe for safe
+  // multi-line messages. Hooks run; commit.gpgsign honored.
   try {
-    const args = ["commit", "--no-verify", "-F", "-"];
+    const args = ["commit", "-F", "-"];
     if (options?.allowEmpty) args.push("--allow-empty");
     const result = execFileSync("git", args, {
       cwd: basePath,

package/src/resources/extensions/gsd/notifications.ts

@@ -4,7 +4,6 @@
 import { execFileSync } from "node:child_process";
 import type { NotificationPreferences } from "./types.js";
 import { loadEffectiveGSDPreferences } from "./preferences.js";
-import { CmuxClient, emitOsc777Notification, resolveCmuxConfig } from "../cmux/index.js";
 import { sendRemoteNotification } from "../remote-questions/notify.js";
 
 export type NotifyLevel = "info" | "success" | "warning" | "error";
@@ -39,20 +38,33 @@ export function sendDesktopNotification(
 
   if (!shouldSendDesktopNotification(kind, loaded?.notifications)) return;
 
-  const cmux = resolveCmuxConfig(loaded);
-  if (cmux.notifications) {
-    const delivered = CmuxClient.fromPreferences(loaded).notify(title, message);
-    if (delivered) return;
-    emitOsc777Notification(title, message);
-  }
-
-  try {
-    const command = buildDesktopNotificationCommand(process.platform, title, message, level);
-    if (!command) return;
-    execFileSync(command.file, command.args, { timeout: 3000, stdio: "ignore" });
-  } catch {
-    // Non-fatal — desktop notifications are best-effort
-  }
+  // cmux delivery and desktop delivery are independent — if cmux import or
+  // delivery fails, we must still attempt the native desktop notification.
+  const runCmux = async () => {
+    try {
+      const { CmuxClient, emitOsc777Notification, resolveCmuxConfig } = await import("../cmux/index.js");
+      const cmux = resolveCmuxConfig(loaded);
+      if (cmux.notifications) {
+        const delivered = CmuxClient.fromPreferences(loaded).notify(title, message);
+        if (delivered) return true;
+        emitOsc777Notification(title, message);
+      }
+    } catch {
+      // cmux unavailable — fall through to desktop notification
+    }
+    return false;
+  };
+
+  void runCmux().then((deliveredByCmux) => {
+    if (deliveredByCmux) return;
+    try {
+      const command = buildDesktopNotificationCommand(process.platform, title, message, level);
+      if (!command) return;
+      execFileSync(command.file, command.args, { timeout: 3000, stdio: "ignore" });
+    } catch {
+      // Non-fatal — desktop notifications are best-effort
+    }
+  }).catch(() => {});
 }
 
 export function shouldSendDesktopNotification(

package/src/resources/extensions/gsd/preferences-validation.ts

@@ -987,6 +987,27 @@ export function validatePreferences(preferences: GSDPreferences): {
     if (g.merge_to_main !== undefined) {
       warnings.push("git.merge_to_main is deprecated — milestone-level merge is now always used. Remove this setting.");
     }
+    // #4765 — collapse cadence + milestone resquash
+    if (g.collapse_cadence !== undefined) {
+      const validCadence = new Set(["milestone", "slice"]);
+      if (typeof g.collapse_cadence === "string" && validCadence.has(g.collapse_cadence)) {
+        git.collapse_cadence = g.collapse_cadence as "milestone" | "slice";
+      } else {
+        errors.push("git.collapse_cadence must be one of: milestone, slice");
+      }
+    }
+    if (g.milestone_resquash !== undefined) {
+      if (typeof g.milestone_resquash === "boolean") {
+        git.milestone_resquash = g.milestone_resquash;
+        const cadence = (git.collapse_cadence as string | undefined)
+          ?? (typeof g.collapse_cadence === "string" ? g.collapse_cadence : undefined);
+        if (cadence !== "slice") {
+          warnings.push('git.milestone_resquash is ignored unless git.collapse_cadence is "slice"');
+        }
+      } else {
+        errors.push("git.milestone_resquash must be a boolean");
+      }
+    }
 
     if (Object.keys(git).length > 0) {
       validated.git = git as GitPreferences;

package/src/resources/extensions/gsd/prompt-cache-optimizer.ts

@@ -55,6 +55,10 @@ const SEMI_STATIC_LABELS = new Set([
   "prior-summaries",
   "project-context",
   "overrides",
+  // KNOWLEDGE is milestone-scoped (stable within a session), so it belongs
+  // in the cacheable prefix. See issue #4719.
+  "knowledge",
+  "project-knowledge",
 ]);
 
 /** Labels that change per-task */

package/src/resources/extensions/gsd/prompts/complete-milestone.md

@@ -10,14 +10,18 @@ Your working directory is `{{workingDirectory}}`. All file reads, writes, and sh
 
 All slices are done. You are closing out the milestone — verifying that the assembled work actually delivers the promised outcome, writing the milestone summary, and updating project state. The milestone summary is the final record. After you finish, the system merges the worktree back to the integration branch. If there are queued milestones, the next one starts its own research → plan → execute cycle from a clean slate — the milestone summary is how it learns what was already built.
 
-All relevant context has been preloaded below — the roadmap, all slice summaries, requirements, decisions, and project context are inlined. Start working immediately without re-reading these files.
+Preloaded context below — the roadmap, compact slice-summary excerpts, requirements, decisions, and project context. **Slice summaries are excerpts, not full files.** They include frontmatter fields, section heads (deviations, known limitations, follow-ups), and short narrative only. When drafting LEARNINGS, the Decision Re-evaluation table, or cross-slice narrative, Read the full slice SUMMARY.md files listed under "On-demand Slice Summaries" — do that selectively as needed, not preemptively.
+
+Start with what the excerpts give you. Read full files when the section heads signal richer context you need.
+
+**On-demand Read ordering:** Complete all slice SUMMARY Reads you need for cross-slice synthesis, the Decision Re-evaluation table, and LEARNINGS **before** calling `gsd_complete_milestone` (step 10). Once that tool runs, the milestone is marked complete in the DB — running out of tool budget between step 10 and the LEARNINGS write (step 12) leaves the milestone committed without its LEARNINGS artifact.
 
 {{inlinedContext}}
 
 Then:
 1. Use the **Milestone Summary** output template from the inlined context above
 2. {{skillActivation}}
-3. **Verify code changes exist.** Run `git diff --stat HEAD $(git merge-base HEAD main) -- ':!.gsd/'` (or the equivalent for the integration branch). If no non-`.gsd/` files appear in the diff, the milestone produced only planning artifacts and no actual code. Record this as a **verification failure**.
+3. **Verify code changes exist.** Compare the milestone against its integration branch (usually `main`, `master`, or the recorded integration branch), using the merge-base as the older revision and `HEAD` as the newer revision. If that branch diff lists non-`.gsd/` files, code-change verification passes. If `HEAD` is already the same commit as the integration branch/merge-base (a retry-on-main self-diff), do **not** treat the empty branch diff as proof of missing code. Instead, inspect milestone-scoped commit evidence such as recent commits with `GSD-Unit: {{milestoneId}}` or production `GSD-Task: Sxx/Tyy` trailers whose diff also touches `.gsd/milestones/{{milestoneId}}/`, then check those commits for non-`.gsd/` files. Only record a **verification failure** when neither the branch diff nor milestone-scoped commit evidence shows implementation files.
 4. Verify each **success criterion** from the milestone definition in `{{roadmapPath}}`. For each criterion, confirm it was met with specific evidence from slice summaries, test results, or observable behavior. Record any criterion that was NOT met as a **verification failure**.
 5. Verify the milestone's **definition of done** — all slices are `[x]`, all slice summaries exist, and any cross-slice integration points work correctly. Record any unmet items as a **verification failure**.
 6. If the roadmap includes a **Horizontal Checklist**, verify each item was addressed during the milestone. Note unchecked items in the milestone summary.

package/src/resources/extensions/gsd/prompts/discuss-headless.md

@@ -164,9 +164,18 @@ Preserve the specification's exact terminology, emphasis, and specific framing.
 
 ### Ready-phrase pre-condition (NON-BYPASSABLE)
 
-Before emitting the ready phrase, verify in the CURRENT turn that you have written `.gsd/PROJECT.md`, `.gsd/REQUIREMENTS.md`, `{{contextPath}}`, and called `gsd_plan_milestone`. If any is missing, **STOP** — emit the missing tool calls in this same turn. The system rejects premature ready signals and retries are capped.
+Before emitting the ready phrase, verify in the CURRENT turn that you have:
 
-After writing the files, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
+- [ ] Written `.gsd/PROJECT.md` (step 2)
+- [ ] Written `.gsd/REQUIREMENTS.md` (step 3)
+- [ ] Written `{{contextPath}}` (step 4)
+- [ ] Called `gsd_plan_milestone` (step 5)
+
+If ANY box is unchecked, **STOP**. Do NOT emit the ready phrase. Emit the missing tool calls in this same turn. The system detects missing artifacts and will reject premature ready signals — you will be asked again and retries are capped.
+
+Do not announce the ready phrase as something you are "about to" do. Do not narrate "now writing the files" as a substitute for actually writing them. The ready phrase is a post-write signal, not an intent signal.
+
+After completing steps 1–7 above, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
 
 ### Multi-Milestone
 
@@ -240,9 +249,19 @@ For single-milestone projects, do NOT write this file.
 
 ### Ready-phrase pre-condition (NON-BYPASSABLE)
 
-Before emitting the ready phrase, verify in the CURRENT turn that you have written `.gsd/PROJECT.md`, `.gsd/REQUIREMENTS.md`, the primary `CONTEXT.md`, called `gsd_plan_milestone` for the primary milestone, and written `.gsd/DISCUSSION-MANIFEST.json` with `gates_completed === total`. If any is missing, **STOP** — emit the missing tool calls in this same turn. The system rejects premature ready signals and retries are capped.
+Before emitting the ready phrase, verify in the CURRENT turn that you have:
+
+- [ ] Written `.gsd/PROJECT.md`
+- [ ] Written `.gsd/REQUIREMENTS.md`
+- [ ] Written the primary milestone `CONTEXT.md`
+- [ ] Called `gsd_plan_milestone` for the primary milestone
+- [ ] Written `.gsd/DISCUSSION-MANIFEST.json` with `gates_completed === total`
+
+If ANY box is unchecked, **STOP**. Do NOT emit the ready phrase. Emit the missing tool calls in this same turn. The system detects missing artifacts and will reject premature ready signals — you will be asked again and retries are capped.
+
+Do not announce the ready phrase as something you are "about to" do. Do not narrate "now writing the files" as a substitute for actually writing them. The ready phrase is a post-write signal, not an intent signal.
 
-After writing the files, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
+After completing every step above, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
 
 ## Critical Rules
 

package/src/resources/extensions/gsd/prompts/doctor-heal.md

@@ -7,9 +7,10 @@ Rules:
 2. Read before edit.
 3. Prefer fixing authoritative artifacts over masking warnings.
 4. For missing summaries or UAT files, generate the real artifact from existing slice/task context when possible — do not leave placeholders if you can reconstruct the real content.
-5. After each repair cluster, verify the relevant invariant directly from disk.
-6. When done, rerun `/gsd doctor {{doctorCommandSuffix}}` mentally by ensuring the remaining issue set for this scope is reduced or cleared.
-7. Do NOT query `.gsd/gsd.db` directly via `sqlite3` or `node -e require('better-sqlite3')` — use `gsd_milestone_status` to inspect DB state. Direct access bypasses the WAL connection owned by the engine and can corrupt in-flight writes.
+5. For a missing milestone `CONTEXT.md` when the milestone is already past `pre-planning` (phase is `executing`, `summarizing`, `validating-milestone`, or `completing-milestone`): the artifact was skipped during bootstrap and must be reconstructed before execution can resume. Read `PROJECT.md`, `REQUIREMENTS.md`, the milestone's `ROADMAP.md`, and any slice-level context on disk, then write `.gsd/milestones/<milestone-id>/<milestone-id>-CONTEXT.md` with the real context for the scoped milestone. Do not leave a stub — the plan gate will reject it on the next cycle.
+6. After each repair cluster, verify the relevant invariant directly from disk.
+7. When done, rerun `/gsd doctor {{doctorCommandSuffix}}` mentally by ensuring the remaining issue set for this scope is reduced or cleared.
+8. Do NOT query `.gsd/gsd.db` directly via `sqlite3` or `node -e require('better-sqlite3')` — use `gsd_milestone_status` to inspect DB state. Direct access bypasses the WAL connection owned by the engine and can corrupt in-flight writes.
 
 ## Doctor Summary
 
@@ -27,4 +28,4 @@ Then:
 - Repair the unresolved issues in scope
 - Keep changes minimal and targeted
 - If unresolved issues remain outside scope, leave them untouched and mention them briefly
-- End with: "GSD doctor heal complete."
+- End with: "GSD doctor heal complete."

package/src/resources/extensions/gsd/prompts/plan-slice.md

@@ -20,9 +20,22 @@ Pay particular attention to **Forward Intelligence** sections — they contain h
 
 You have full tool access. Before decomposing, explore the relevant code to ground your plan in reality.
 
-### Verify Roadmap Assumptions
+### Verify Roadmap Assumptions (JIT Reassessment — ADR-003 §4)
 
-Check prior slice summaries (inlined above as dependency summaries, if present). If prior slices discovered constraints, changed approaches, or flagged fragility, adjust your plan accordingly. The roadmap description may be stale — verify it against the current codebase state.
+Before planning this slice, verify that the roadmap's assumptions still hold given prior slice summaries. Check inlined dependency summaries (below) for discovered constraints, changed approaches, or flagged fragility.
+
+**If the remaining roadmap needs adjustment, modify it before proceeding:**
+
+- If a downstream slice's title/demo/dependencies are now wrong, call `gsd_reassess_roadmap` with the corrected `sliceChanges.modified` entry.
+- If new work surfaced that deserves its own slice, add it via `sliceChanges.added`.
+- If a downstream slice is now redundant or out of scope, remove it via `sliceChanges.removed`.
+- **Bias strongly toward "roadmap is fine."** Most slice completions produce no structural change. Only adjust when there is concrete evidence a downstream slice is wrong — not speculative concern. Over-reassessment is costlier than a later mid-slice replan.
+
+Completed slices are immutable: never modify or remove a slice whose status is complete.
+
+Then proceed with planning this slice against the (possibly updated) roadmap.
+
+The roadmap description may be stale — verify it against the current codebase state.
 
 ### Explore Slice Scope
 

package/src/resources/extensions/gsd/prompts/system.md

@@ -213,6 +213,7 @@ Fix the root cause, not symptoms. When applying a temporary mitigation, label it
 - State uncertainty plainly: "Not sure this handles X - testing it." No performed confidence, no hedging paragraphs.
 - All user-visible narration must be grammatical English. Do not emit compressed planner notes like "Need inspect X" or "Maybe read Y first". If it would look acceptable in a commit comment or standup note, it's acceptable here.
 - When debugging, stay curious. Problems are puzzles. Say what's interesting about the failure before reaching for fixes.
+- After completing a task, give a brief completion summary and present 2-4 contextual next-step options as a numbered list (the last option is always "Other"). This reduces cognitive load by letting the user pick rather than formulate what's next. Omit the numbered list when the response must follow a strict output format (JSON, patches, commit messages, structured data).
 
 Good narration: "Three existing handlers follow a middleware pattern - using that instead of a custom wrapper."
 Good narration: "Tests pass. Running slice-level verification."

package/src/resources/extensions/gsd/reports.ts

@@ -14,8 +14,9 @@
  * Manual: /gsd export --html
  */
 
-import { writeFileSync, readFileSync, mkdirSync, existsSync } from 'node:fs';
+import { readFileSync, mkdirSync, existsSync } from 'node:fs';
 import { join, basename } from 'node:path';
+import { atomicWriteSync } from './atomic-write.js';
 import { gsdRoot } from './paths.js';
 import { formatCost, formatTokenCount } from './metrics.js';
 import { formatDateShort, formatDuration } from '../shared/format-utils.js';
@@ -83,7 +84,7 @@ export function loadReportsIndex(basePath: string): ReportsIndex | null {
 function saveReportsIndex(basePath: string, index: ReportsIndex): void {
   const dir = reportsDir(basePath);
   mkdirSync(dir, { recursive: true });
-  writeFileSync(reportsIndexPath(basePath), JSON.stringify(index, null, 2) + '\n', 'utf-8');
+  atomicWriteSync(reportsIndexPath(basePath), JSON.stringify(index, null, 2) + '\n', 'utf-8');
 }
 
 // ─── Write a report snapshot ──────────────────────────────────────────────────
@@ -121,7 +122,7 @@ export function writeReportSnapshot(args: WriteReportSnapshotArgs): string {
   const filename = `${prefix}-${timestamp}.html`;
   const filePath = join(dir, filename);
 
-  writeFileSync(filePath, args.html, 'utf-8');
+  atomicWriteSync(filePath, args.html, 'utf-8');
 
   // Load or init registry
   const existing = loadReportsIndex(args.basePath);
@@ -170,7 +171,7 @@ export function writeReportSnapshot(args: WriteReportSnapshotArgs): string {
 
 export function regenerateHtmlIndex(basePath: string, index: ReportsIndex): void {
   const html = buildIndexHtml(index);
-  writeFileSync(reportsHtmlIndexPath(basePath), html, 'utf-8');
+  atomicWriteSync(reportsHtmlIndexPath(basePath), html, 'utf-8');
 }
 
 function buildIndexHtml(index: ReportsIndex): string {

package/src/resources/extensions/gsd/safety/git-checkpoint.ts

@@ -72,6 +72,21 @@ export function rollbackToCheckpoint(
       return false;
     }
 
+    // Preserve any staged or untracked user work before the hard reset.
+    // The user may have a partial fix staged that they wanted to inspect;
+    // reset --hard wipes both staged and unstaged changes (reflog only
+    // covers committed state). Push a labeled stash first so recovery
+    // is possible. (Issue #4980 HIGH-4)
+    try {
+      execFileSync(
+        "git",
+        ["stash", "push", "--include-untracked", "-m", `gsd: pre-rollback-stash ${unitId} ${new Date().toISOString()}`],
+        { cwd: basePath, stdio: ["ignore", "pipe", "pipe"], encoding: "utf-8" },
+      );
+    } catch {
+      /* nothing to stash, or stash refused — proceed with reset */
+    }
+
     // Reset branch pointer and working tree to checkpoint SHA in one step.
     // Using `git reset --hard <sha>` works on the currently checked-out branch
     // (unlike `git branch -f` which is rejected for checked-out branches).

package/src/resources/extensions/gsd/service-tier.ts

@@ -2,8 +2,8 @@
  * Service Tier — gating, status formatting, icon resolution, and
  * the /gsd fast command handler.
  *
- * Service tiers (priority/flex) are an OpenAI feature that only applies
- * to gpt-5.4 variants. This module centralizes the model-gating logic
+ * Service tiers (priority/flex) are an OpenAI feature that currently only
+ * applies to gpt-5.4 variants in GSD. This module centralizes the model-gating logic
  * so that icons, preferences, and the before_provider_request hook all
  * use a single source of truth.
  */
@@ -35,6 +35,9 @@ const SERVICE_TIER_SCOPE_NOTE = "Only affects gpt-5.4 models, regardless of prov
  * (set via CAPABILITY_PATCHES in packages/pi-ai/src/models.ts). When callers
  * have access to the full Model object, prefer reading capabilities directly.
  *
+ * GPT-5.5 is intentionally excluded until we verify its provider payload
+ * contract instead of assuming `service_tier` support.
+ *
  * See: https://github.com/gsd-build/gsd-2/issues/2546
  */
 const SERVICE_TIER_MODEL_PREFIXES = ["gpt-5.4"] as const;

package/src/resources/extensions/gsd/session-lock.ts

@@ -188,13 +188,14 @@ function ensureExitHandler(_gsdDir: string): void {
     // Lock files accumulate across main project .gsd/, worktree .gsd/,
     // and projects registry paths — cleanup must cover all of them.
     for (const dir of _lockDirRegistry) {
+      const lockFile = join(dir, LOCK_FILE);
+      const ownsRegisteredLock = isLockFileOwnedByCurrentProcess(lockFile);
       try {
-        const lockFile = join(dir, LOCK_FILE);
-        if (existsSync(lockFile)) unlinkSync(lockFile);
+        if (ownsRegisteredLock && existsSync(lockFile)) unlinkSync(lockFile);
       } catch { /* best-effort */ }
       try {
         const lockDir = join(dir + ".lock");
-        if (existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
+        if (ownsRegisteredLock && existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
       } catch { /* best-effort */ }
     }
   });
@@ -526,10 +527,13 @@ export function releaseSessionLock(basePath: string): void {
     _releaseFunction = null;
   }
 
-  // Remove the lock file at the current path
+  // Remove the lock file at the current path only if it still belongs to us.
+  // Lost-lock cleanup can run after another process has taken ownership; in
+  // that case deleting auto.lock would erase the newer owner's evidence.
   const lp = lockPath(basePath);
+  const ownsPrimaryLock = isLockFileOwnedByCurrentProcess(lp);
   try {
-    if (existsSync(lp)) unlinkSync(lp);
+    if (ownsPrimaryLock && existsSync(lp)) unlinkSync(lp);
   } catch {
     // Non-fatal
   }
@@ -540,12 +544,12 @@ export function releaseSessionLock(basePath: string): void {
   const lockTarget = effectiveLockTarget(gsdDir);
   try {
     const lockDir = join(lockTarget + ".lock");
-    if (existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
+    if (ownsPrimaryLock && existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
   } catch {
     // Non-fatal
   }
   // Also clean the per-milestone parallel directory itself if it exists
-  if (lockTarget !== gsdDir) {
+  if (ownsPrimaryLock && lockTarget !== gsdDir) {
     try {
       if (existsSync(lockTarget)) rmSync(lockTarget, { recursive: true, force: true });
     } catch {
@@ -556,13 +560,14 @@ export function releaseSessionLock(basePath: string): void {
   // Clean ALL registered lock paths (#1578) — lock files accumulate across
   // main project .gsd/, worktree .gsd/, and projects registry paths.
   for (const dir of _lockDirRegistry) {
+    const lockFile = join(dir, LOCK_FILE);
+    const ownsRegisteredLock = isLockFileOwnedByCurrentProcess(lockFile);
     try {
-      const lockFile = join(dir, LOCK_FILE);
-      if (existsSync(lockFile)) unlinkSync(lockFile);
+      if (ownsRegisteredLock && existsSync(lockFile)) unlinkSync(lockFile);
     } catch { /* best-effort */ }
     try {
       const lockDir = join(dir + ".lock");
-      if (existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
+      if (ownsRegisteredLock && existsSync(lockDir)) rmSync(lockDir, { recursive: true, force: true });
     } catch { /* best-effort */ }
   }
   _lockDirRegistry.clear();
@@ -619,6 +624,11 @@ function readExistingLockData(lp: string): SessionLockData | null {
   }
 }
 
+function isLockFileOwnedByCurrentProcess(lp: string): boolean {
+  const existing = readExistingLockData(lp);
+  return existing?.pid === process.pid;
+}
+
 /**
  * Retry-tolerant variant of readExistingLockData for use in onCompromised and
  * other paths where a transient filesystem hiccup (NFS/CIFS latency, macOS APFS