gsd-pi 2.76.0-dev.82e249f7b → 2.76.0-dev.97f5583d9

package/src/resources/extensions/gsd/pre-execution-checks.ts

@@ -91,8 +91,13 @@ export function extractPackageReferences(description: string): string[] {
     }
   }
 
-  // require('pkg') or import from 'pkg' in code blocks
-  const importPattern = /(?:require\s*\(\s*['"]|from\s+['"])([a-zA-Z0-9@/_-]+)['"\)]/g;
+  // require('pkg') or `import ... from 'pkg'` in code blocks.
+  // The `from\s+['"]` branch MUST be preceded by an `import` keyword so that
+  // natural-language prose like `from "What's Next"` or `from 'master'` does
+  // not produce false package-existence failures.  Requiring the leading import
+  // keyword anchors the match to JavaScript/TypeScript syntax.
+  // See: https://github.com/gsd-build/gsd-2/issues/4388
+  const importPattern = /(?:require\s*\(\s*['"]|import\b[\s\S]*?\bfrom\s+['"])([a-zA-Z0-9@/_-]+)['"\)]/g;
   let importMatch: RegExpExecArray | null;
   while ((importMatch = importPattern.exec(description)) !== null) {
     // Skip relative imports and node builtins
@@ -325,7 +330,12 @@ function extractPathFromAnnotation(raw: string): string {
 
   const annotatedMatch = trimmed.match(/^(.+?)\s+[—–-]\s+.+$/);
   if (annotatedMatch) {
-    return annotatedMatch[1].trim();
+    const prefix = annotatedMatch[1].trim();
+    const prefixBacktickMatch = prefix.match(/`([^`]+)`/);
+    if (prefixBacktickMatch && looksLikePathOrUrl(prefixBacktickMatch[1].trim())) {
+      return prefixBacktickMatch[1].trim();
+    }
+    return prefix.replace(/`/g, "").trim();
   }
 
   // Fallback: scan all backticked tokens and return the first one that looks
@@ -388,13 +398,19 @@ function containsGlobPattern(candidate: string): boolean {
 
 /**
  * Build a set of files that will be created by tasks up to (but not including) taskIndex.
+ * Also includes outputs of completed tasks at any position — a completed task has already
+ * run and its outputs are available regardless of sequence position or disk state (#4071).
  * All paths are normalized for consistent comparison.
  */
 function getExpectedOutputsUpTo(tasks: TaskRow[], taskIndex: number): Set<string> {
   const outputs = new Set<string>();
-  for (let i = 0; i < taskIndex; i++) {
-    for (const file of tasks[i].expected_output) {
-      outputs.add(normalizeFilePath(file));
+  for (let i = 0; i < tasks.length; i++) {
+    const task = tasks[i];
+    // Include prior tasks (i < taskIndex) OR completed tasks at any position
+    if (i < taskIndex || task.status === "completed") {
+      for (const file of task.expected_output) {
+        outputs.add(normalizeFilePath(file));
+      }
     }
   }
   return outputs;
@@ -481,13 +497,19 @@ export function checkTaskOrdering(
   const results: PreExecutionCheckJSON[] = [];
 
   // Build map: normalized file → task index that creates it
-  const fileCreators = new Map<string, { taskId: string; index: number; originalPath: string }>();
+  const fileCreators = new Map<string, { taskId: string; index: number; originalPath: string; completed: boolean }>();
   for (let i = 0; i < tasks.length; i++) {
     const task = tasks[i];
     for (const file of task.expected_output) {
       const normalizedFile = normalizeFilePath(file);
-      if (!fileCreators.has(normalizedFile)) {
-        fileCreators.set(normalizedFile, { taskId: task.id, index: i, originalPath: file });
+      const existing = fileCreators.get(normalizedFile);
+      if (!existing || (!existing.completed && task.status === "completed")) {
+        fileCreators.set(normalizedFile, {
+          taskId: task.id,
+          index: i,
+          originalPath: file,
+          completed: task.status === "completed",
+        });
       }
     }
   }
@@ -511,7 +533,11 @@ export function checkTaskOrdering(
       const creator = fileCreators.get(normalizedFile);
       const absolutePath = resolve(basePath, normalizedFile);
       const existsOnDisk = existsSync(absolutePath);
-      if (creator && creator.index > i && !existsOnDisk) {
+      // Skip if the creating task has already completed — its output is available
+      // regardless of disk state (e.g. file was a temp artifact cleaned up after
+      // the task ran, or a replan introduced a new earlier-sequence task that
+      // reads this pre-execution output). (#4071)
+      if (creator && creator.index > i && !existsOnDisk && !creator.completed) {
         // Task reads file that is created later — impossible ordering
         results.push({
           category: "file",

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

@@ -28,6 +28,37 @@ export interface ContextManagementConfig {
   compaction_threshold_percent?: number;  // default: 0.70, range: 0.5-0.95
   tool_result_max_chars?: number;         // default: 800, range: 200-10000
 }
+
+/**
+ * Opt-in tool-output sandboxing for sub-sessions. When enabled, the gsd_exec
+ * MCP tool runs scripts in an isolated subprocess and returns only a short
+ * digest to the calling agent's context window; full stdout/stderr persist
+ * in the project memory store and can be retrieved by id later.
+ *
+ * Inspired by mksglu/context-mode (Elastic License 2.0). This is an
+ * independent implementation — no upstream code is incorporated.
+ */
+export interface ContextModeConfig {
+  /** Master switch. Default: true (opt-out via `enabled: false`). */
+  enabled?: boolean;
+  /** Per-invocation timeout in milliseconds. Default: 30_000. Range: 1_000–600_000. */
+  exec_timeout_ms?: number;
+  /** Cap on persisted stdout bytes per invocation. Default: 1_048_576 (1 MiB). Range: 4_096–16_777_216. */
+  exec_stdout_cap_bytes?: number;
+  /** Number of trailing stdout characters returned in the digest. Default: 300. Range: 0–4_000. */
+  exec_digest_chars?: number;
+  /** Environment variables forwarded to sandboxed processes (case-sensitive names). PATH and HOME are always forwarded. */
+  exec_env_allowlist?: string[];
+}
+
+/**
+ * Resolve whether context-mode features (gsd_exec sandbox + compaction
+ * snapshot) should be active. Default is ON: missing config or missing
+ * `enabled` is treated as true. Only `enabled: false` disables.
+ */
+export function isContextModeEnabled(prefs: { context_mode?: ContextModeConfig } | null | undefined): boolean {
+  return prefs?.context_mode?.enabled !== false;
+}
 import type { GitHubSyncConfig } from "../github-sync/types.js";
 
 // ─── Workflow Modes ──────────────────────────────────────────────────────────
@@ -117,6 +148,7 @@ export const KNOWN_PREFERENCE_KEYS = new Set<string>([
   "flat_rate_providers",
   "language",
   "context_window_override",
+  "context_mode",
 ]);
 
 /** Canonical list of all dispatch unit types. */
@@ -300,6 +332,12 @@ export interface GSDPreferences {
    */
   context_window_override?: number;
   context_management?: ContextManagementConfig;
+  /**
+   * Tool-output sandboxing via gsd_exec. Keeps sub-session context windows
+   * clean by running scripts in a subprocess and only surfacing a short
+   * digest. See `ContextModeConfig`. Default: disabled.
+   */
+  context_mode?: ContextModeConfig;
   token_profile?: TokenProfile;
   phases?: PhaseSkipPreferences;
   auto_visualize?: boolean;
@@ -354,6 +392,14 @@ export interface GSDPreferences {
     checkpoints?: boolean;
     auto_rollback?: boolean;
     timeout_scale_cap?: number;
+    /**
+     * Glob patterns for files that are always expected side-effects of any task.
+     * Files matching any pattern here are excluded from unexpected-change warnings.
+     * Supports standard glob syntax (e.g. `tracking/history/**`, `*.log`).
+     * Fixes #4385/#4436 — audit-trail snapshots, build artifacts, and other
+     * project-level secondary writes shouldn't require per-task declaration.
+     */
+    file_change_allowlist?: string[];
   };
 
 

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

@@ -644,6 +644,50 @@ export function validatePreferences(preferences: GSDPreferences): {
     }
   }
 
+  // ─── Context Mode (gsd_exec sandbox) ────────────────────────────────────
+  if (preferences.context_mode !== undefined) {
+    if (typeof preferences.context_mode === "object" && preferences.context_mode !== null) {
+      const cmode = preferences.context_mode as unknown as Record<string, unknown>;
+      const validCmode: Record<string, unknown> = {};
+
+      if (cmode.enabled !== undefined) {
+        if (typeof cmode.enabled === "boolean") validCmode.enabled = cmode.enabled;
+        else errors.push("context_mode.enabled must be a boolean");
+      }
+      if (cmode.exec_timeout_ms !== undefined) {
+        const t = cmode.exec_timeout_ms;
+        if (typeof t === "number" && t >= 1000 && t <= 600_000) validCmode.exec_timeout_ms = Math.floor(t);
+        else errors.push("context_mode.exec_timeout_ms must be a number between 1000 and 600000");
+      }
+      if (cmode.exec_stdout_cap_bytes !== undefined) {
+        const b = cmode.exec_stdout_cap_bytes;
+        if (typeof b === "number" && b >= 4096 && b <= 16_777_216) validCmode.exec_stdout_cap_bytes = Math.floor(b);
+        else errors.push("context_mode.exec_stdout_cap_bytes must be a number between 4096 and 16777216");
+      }
+      if (cmode.exec_digest_chars !== undefined) {
+        const c = cmode.exec_digest_chars;
+        if (typeof c === "number" && c >= 0 && c <= 4000) validCmode.exec_digest_chars = Math.floor(c);
+        else errors.push("context_mode.exec_digest_chars must be a number between 0 and 4000");
+      }
+      if (cmode.exec_env_allowlist !== undefined) {
+        if (
+          Array.isArray(cmode.exec_env_allowlist) &&
+          cmode.exec_env_allowlist.every((v) => typeof v === "string" && /^[A-Z_][A-Z0-9_]*$/i.test(v))
+        ) {
+          validCmode.exec_env_allowlist = cmode.exec_env_allowlist;
+        } else {
+          errors.push("context_mode.exec_env_allowlist must be an array of valid env var names");
+        }
+      }
+
+      if (Object.keys(validCmode).length > 0) {
+        validated.context_mode = validCmode as any;
+      }
+    } else {
+      errors.push("context_mode must be an object");
+    }
+  }
+
   // ─── Parallel Config ────────────────────────────────────────────────────
   if (preferences.parallel && typeof preferences.parallel === "object") {
     const p = preferences.parallel as unknown as Record<string, unknown>;
@@ -697,6 +741,41 @@ export function validatePreferences(preferences: GSDPreferences): {
     }
   }
 
+  // ─── Slice Parallel Config ───────────────────────────────────────────────
+  if (preferences.slice_parallel !== undefined) {
+    if (typeof preferences.slice_parallel === "object" && preferences.slice_parallel !== null) {
+      const sp = preferences.slice_parallel as Record<string, unknown>;
+      const validSp: NonNullable<GSDPreferences["slice_parallel"]> = {};
+
+      if (sp.enabled !== undefined) {
+        if (typeof sp.enabled === "boolean") validSp.enabled = sp.enabled;
+        else errors.push("slice_parallel.enabled must be a boolean");
+      }
+
+      if (sp.max_workers !== undefined) {
+        const maxWorkers = typeof sp.max_workers === "number" ? sp.max_workers : Number(sp.max_workers);
+        if (Number.isFinite(maxWorkers) && maxWorkers >= 1 && maxWorkers <= 8) {
+          validSp.max_workers = Math.floor(maxWorkers);
+        } else {
+          errors.push("slice_parallel.max_workers must be a number between 1 and 8");
+        }
+      }
+
+      const knownSliceParallelKeys = new Set(["enabled", "max_workers"]);
+      for (const key of Object.keys(sp)) {
+        if (!knownSliceParallelKeys.has(key)) {
+          warnings.push(`unknown slice_parallel key "${key}" — ignored`);
+        }
+      }
+
+      if (Object.keys(validSp).length > 0) {
+        validated.slice_parallel = validSp;
+      }
+    } else {
+      errors.push("slice_parallel must be an object");
+    }
+  }
+
   // ─── Reactive Execution ─────────────────────────────────────────────────
   if (preferences.reactive_execution !== undefined) {
     if (typeof preferences.reactive_execution === "object" && preferences.reactive_execution !== null) {

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

@@ -68,13 +68,13 @@ export { resolveAllSkillReferences } from "./preferences-skills.js";
 // These lived in preferences-skills.ts but imported loadEffectiveGSDPreferences
 // back from this file, creating a circular dependency. Moved here since they
 // are trivial wrappers over loadEffectiveGSDPreferences.
-export function resolveSkillDiscoveryMode(): SkillDiscoveryMode {
-  const prefs = loadEffectiveGSDPreferences();
+export function resolveSkillDiscoveryMode(basePath?: string): SkillDiscoveryMode {
+  const prefs = loadEffectiveGSDPreferences(basePath);
   return prefs?.preferences.skill_discovery ?? "suggest";
 }
 
-export function resolveSkillStalenessDays(): number {
-  const prefs = loadEffectiveGSDPreferences();
+export function resolveSkillStalenessDays(basePath?: string): number {
+  const prefs = loadEffectiveGSDPreferences(basePath);
   return prefs?.preferences.skill_staleness_days ?? 60;
 }
 
@@ -109,16 +109,16 @@ function legacyGlobalPreferencesPath(): string {
   return join(homedir(), ".pi", "agent", "gsd-preferences.md");
 }
 
-function projectPreferencesPath(): string {
-  return join(gsdRoot(process.cwd()), "PREFERENCES.md");
+function projectPreferencesPath(basePath: string = process.cwd()): string {
+  return join(gsdRoot(basePath), "PREFERENCES.md");
 }
 // Legacy lowercase files can still exist in older projects. Keep them as a
 // compatibility-only fallback, but route new reads/writes through PREFERENCES.md.
 function legacyGlobalPreferencesPathLowercase(): string {
   return join(gsdHome(), "preferences.md");
 }
-function legacyProjectPreferencesPathLowercase(): string {
-  return join(gsdRoot(process.cwd()), "preferences.md");
+function legacyProjectPreferencesPathLowercase(basePath: string = process.cwd()): string {
+  return join(gsdRoot(basePath), "preferences.md");
 }
 
 export function getGlobalGSDPreferencesPath(): string {
@@ -129,8 +129,8 @@ export function getLegacyGlobalGSDPreferencesPath(): string {
   return legacyGlobalPreferencesPath();
 }
 
-export function getProjectGSDPreferencesPath(): string {
-  return projectPreferencesPath();
+export function getProjectGSDPreferencesPath(basePath?: string): string {
+  return projectPreferencesPath(basePath);
 }
 
 // ─── Loading ────────────────────────────────────────────────────────────────
@@ -141,14 +141,14 @@ export function loadGlobalGSDPreferences(): LoadedGSDPreferences | null {
     ?? loadPreferencesFile(legacyGlobalPreferencesPath(), "global");
 }
 
-export function loadProjectGSDPreferences(): LoadedGSDPreferences | null {
-  return loadPreferencesFile(projectPreferencesPath(), "project")
-    ?? loadPreferencesFile(legacyProjectPreferencesPathLowercase(), "project");
+export function loadProjectGSDPreferences(basePath?: string): LoadedGSDPreferences | null {
+  return loadPreferencesFile(projectPreferencesPath(basePath), "project")
+    ?? loadPreferencesFile(legacyProjectPreferencesPathLowercase(basePath), "project");
 }
 
-export function loadEffectiveGSDPreferences(): LoadedGSDPreferences | null {
+export function loadEffectiveGSDPreferences(basePath?: string): LoadedGSDPreferences | null {
   const globalPreferences = loadGlobalGSDPreferences();
-  const projectPreferences = loadProjectGSDPreferences();
+  const projectPreferences = loadProjectGSDPreferences(basePath);
 
   if (!globalPreferences && !projectPreferences) return null;
 
@@ -603,8 +603,8 @@ export function resolvePreDispatchHooks(): PreDispatchHookConfig[] {
  * Worktree isolation requires explicit opt-in because it depends on git
  * branch infrastructure that must be set up before use.
  */
-export function getIsolationMode(): "none" | "worktree" | "branch" {
-  const prefs = loadEffectiveGSDPreferences()?.preferences?.git;
+export function getIsolationMode(basePath?: string): "none" | "worktree" | "branch" {
+  const prefs = loadEffectiveGSDPreferences(basePath)?.preferences?.git;
   if (prefs?.isolation === "worktree") return "worktree";
   if (prefs?.isolation === "branch") return "branch";
   return "none"; // default — no isolation, work on current branch

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

@@ -162,6 +162,10 @@ Preserve the specification's exact terminology, emphasis, and specific framing.
 6. For each architectural or pattern decision, call `gsd_decision_save` — the tool auto-assigns IDs and regenerates `.gsd/DECISIONS.md` automatically.
 7. {{commitInstruction}}
 
+### 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.
+
 After writing the files, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
 
 ### Multi-Milestone
@@ -234,6 +238,10 @@ For single-milestone projects, do NOT write this file.
 
 7. {{multiMilestoneCommitInstruction}}
 
+### 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.
+
 After writing the files, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
 
 ## Critical Rules

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

@@ -339,7 +339,20 @@ These sections are in addition to whatever other context the discussion surfaced
 6. For each architectural or pattern decision made during discussion, call `gsd_decision_save` — the tool auto-assigns IDs and regenerates `.gsd/DECISIONS.md` automatically.
 7. {{commitInstruction}}
 
-After writing the files, say exactly: "Milestone {{milestoneId}} ready." — nothing else. Auto-mode will start automatically.
+### Ready-phrase pre-condition (NON-BYPASSABLE)
+
+Before emitting the ready phrase, verify in the CURRENT turn that you have:
+
+- [ ] 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
 
@@ -418,6 +431,20 @@ For single-milestone projects, do NOT write this file — it is only for multi-m
 
 7. {{multiMilestoneCommitInstruction}}
 
-After writing the files, say exactly: "Milestone M001 ready." — nothing else. Auto-mode will start automatically.
+### Ready-phrase pre-condition (NON-BYPASSABLE)
+
+Before emitting the ready phrase, verify in the CURRENT turn that you have:
+
+- [ ] Written `.gsd/PROJECT.md` (Phase 1)
+- [ ] Written `.gsd/REQUIREMENTS.md` (Phase 1)
+- [ ] Written primary-milestone `CONTEXT.md` (Phase 2)
+- [ ] Called `gsd_plan_milestone` for the primary milestone (Phase 2)
+- [ ] Written `.gsd/DISCUSSION-MANIFEST.json` with `gates_completed === total` (Phase 3)
+
+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 all phases above, say exactly: "Milestone M001 ready." — nothing else. Auto-mode will start automatically.
 
 {{inlinedTemplates}}

package/src/resources/extensions/gsd/prompts/parallel-research-slices.md

@@ -15,8 +15,11 @@ Dispatch ALL slices simultaneously using the `subagent` tool in **parallel mode*
 1. Call `subagent` with `tasks: [...]` containing one entry per slice below
 2. Wait for ALL subagents to complete
 3. Verify each slice's RESEARCH file was written (check the `.gsd/{{mid}}/` directory)
-4. If any subagent failed to write its RESEARCH file, re-run it individually
-5. Report which slices completed research and which (if any) failed
+4. If a subagent failed to write its RESEARCH file, retry it **once** individually
+5. If it fails a second time, write a partial RESEARCH file for that slice with a `## BLOCKER` section explaining the failure — do NOT retry again
+6. Report which slices completed research and which (if any) needed a blocker note
+
+**Important**: Each failed slice gets exactly one retry. After that, write the blocker and move on. Never retry the same slice more than once.
 
 ## Subagent Prompts
 

package/src/resources/extensions/gsd/safety/evidence-collector.ts

@@ -3,10 +3,26 @@
  * Tracks every bash command, file write, and file edit during a unit execution.
  * Evidence is compared against LLM completion claims in evidence-cross-ref.ts.
  *
+ * Evidence is persisted to .gsd/safety/evidence-<mid>-<sid>-<tid>.json so it
+ * survives session restarts (pause/resume, crash recovery). On unit start,
+ * call resetEvidence() then loadEvidenceFromDisk(). On every new tool call,
+ * saveEvidenceToDisk() is called automatically by recordToolCall/recordToolResult.
+ *
  * Follows the same module-level Map pattern as auto-tool-tracking.ts.
  * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
  */
 
+import {
+  existsSync,
+  mkdirSync,
+  readFileSync,
+  writeFileSync,
+  renameSync,
+  unlinkSync,
+} from "node:fs";
+import { join, dirname } from "node:path";
+import { randomBytes } from "node:crypto";
+
 // ─── Types ──────────────────────────────────────────────────────────────────
 
 export interface BashEvidence {
@@ -62,6 +78,109 @@ export function getFilePaths(): string[] {
     .map(e => e.path);
 }
 
+// ─── Persistence (Bug #4385 — evidence must survive session restarts) ────────
+
+/**
+ * Build the path for the evidence JSON file for a given unit.
+ * Lives under .gsd/safety/ which is gitignored and session-scoped.
+ */
+function evidencePath(basePath: string, milestoneId: string, sliceId: string, taskId: string): string {
+  return join(basePath, ".gsd", "safety", `evidence-${milestoneId}-${sliceId}-${taskId}.json`);
+}
+
+/**
+ * Validate that a parsed value is an array of EvidenceEntry objects.
+ * Rejects corrupt / schema-mismatch data rather than letting it poison state.
+ */
+function isEvidenceArray(data: unknown): data is EvidenceEntry[] {
+  if (!Array.isArray(data)) return false;
+  return data.every((e) => {
+    if (e === null || typeof e !== "object") return false;
+    const rec = e as Record<string, unknown>;
+    if (typeof rec.toolCallId !== "string") return false;
+    if (typeof rec.timestamp !== "number") return false;
+    if (rec.kind === "bash") {
+      return (
+        typeof rec.command === "string" &&
+        typeof rec.exitCode === "number" &&
+        typeof rec.outputSnippet === "string"
+      );
+    }
+    if (rec.kind === "write" || rec.kind === "edit") {
+      return typeof rec.path === "string";
+    }
+    return false;
+  });
+}
+
+/**
+ * Persist the current in-memory evidence to disk so it survives a session
+ * restart. Called from saveEvidenceToDisk after recordToolCall/recordToolResult.
+ * Non-fatal — persistence failures must never break unit execution.
+ */
+export function saveEvidenceToDisk(
+  basePath: string,
+  milestoneId: string,
+  sliceId: string,
+  taskId: string,
+): void {
+  try {
+    const path = evidencePath(basePath, milestoneId, sliceId, taskId);
+    mkdirSync(dirname(path), { recursive: true });
+    const tmp = `${path}.tmp.${randomBytes(4).toString("hex")}`;
+    writeFileSync(tmp, JSON.stringify(unitEvidence, null, 2) + "\n", "utf-8");
+    renameSync(tmp, path);
+  } catch {
+    // Non-fatal — don't let persistence failures break unit execution
+  }
+}
+
+/**
+ * Load persisted evidence from disk into the in-memory array.
+ * Call after resetEvidence() on session resume to restore context for a
+ * partially-executed unit. If the file does not exist (fresh unit), this
+ * is a no-op — getEvidence() will return [] which is correct.
+ */
+export function loadEvidenceFromDisk(
+  basePath: string,
+  milestoneId: string,
+  sliceId: string,
+  taskId: string,
+): void {
+  try {
+    const path = evidencePath(basePath, milestoneId, sliceId, taskId);
+    if (!existsSync(path)) return;
+    const raw = readFileSync(path, "utf-8");
+    const parsed = JSON.parse(raw);
+    if (isEvidenceArray(parsed)) {
+      unitEvidence = parsed;
+    }
+  } catch {
+    // Non-fatal — corrupt / missing file is treated as empty evidence
+  }
+}
+
+/**
+ * Delete the persisted evidence file for a unit after it has been fully
+ * processed. Prevents stale evidence from affecting future retries of
+ * the same unit ID.
+ */
+export function clearEvidenceFromDisk(
+  basePath: string,
+  milestoneId: string,
+  sliceId: string,
+  taskId: string,
+): void {
+  try {
+    const path = evidencePath(basePath, milestoneId, sliceId, taskId);
+    if (existsSync(path)) {
+      unlinkSync(path);
+    }
+  } catch {
+    // Non-fatal
+  }
+}
+
 // ─── Recording (called from register-hooks.ts) ─────────────────────────────
 
 /**

package/src/resources/extensions/gsd/safety/file-change-validator.ts

@@ -4,15 +4,23 @@
  *
  * Uses tasks.expected_output (DB column, populated from per-task ## Expected Output)
  * and tasks.files (from slice PLAN.md - Files: subline) as the expected set.
- * Compares against git diff HEAD~1 --name-only after auto-commit.
+ * Compares against `git diff-tree --root --no-commit-id -r --name-only HEAD` after auto-commit.
+ * Using diff-tree --root handles initial commits, shallow clones, and merge commits correctly
+ * (Bug #4385 — git diff HEAD~1 failed on initial commits).
  *
  * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
  */
 
+import { createRequire } from "node:module";
 import { execFileSync } from "node:child_process";
 import { normalizePlannedFileReference } from "../files.js";
 import { logWarning } from "../workflow-logger.js";
 
+const _require = createRequire(import.meta.url);
+type PicomatchMatcher = (input: string) => boolean;
+type PicomatchFn = (pattern: string, opts?: { dot?: boolean }) => PicomatchMatcher;
+const picomatch = _require("picomatch") as PicomatchFn;
+
 // ─── Types ──────────────────────────────────────────────────────────────────
 
 export interface FileViolation {
@@ -43,6 +51,7 @@ export function validateFileChanges(
   basePath: string,
   expectedOutput: string[],
   plannedFiles: string[],
+  fileChangeAllowlist: string[] = [],
 ): FileChangeAudit | null {
   const allExpected = new Set([...expectedOutput, ...plannedFiles]);
 
@@ -63,8 +72,12 @@ export function validateFileChanges(
     ),
   );
 
-  // Compute symmetric difference
-  const unexpectedFiles = projectFiles.filter(f => !normalizedExpected.has(f));
+  // Build allowlist matchers once (dot: true so patterns like `**/.hidden` work).
+  const allowlistMatchers = fileChangeAllowlist.map(p => picomatch(p, { dot: true }));
+  const isAllowlisted = (f: string) => allowlistMatchers.some(m => m(f));
+
+  // Compute symmetric difference, excluding allowlisted files
+  const unexpectedFiles = projectFiles.filter(f => !normalizedExpected.has(f) && !isAllowlisted(f));
   const missingFiles = [...normalizedExpected].filter(f => !projectFiles.includes(f));
 
   const violations: FileViolation[] = [];
@@ -100,7 +113,7 @@ function getChangedFilesFromLastCommit(basePath: string): string[] | null {
   try {
     const result = execFileSync(
       "git",
-      ["diff", "--name-only", "HEAD~1", "HEAD"],
+      ["diff-tree", "--root", "--no-commit-id", "-r", "--name-only", "HEAD"],
       { cwd: basePath, stdio: ["ignore", "pipe", "pipe"], encoding: "utf-8" },
     ).trim();
     return result ? result.split("\n").filter(Boolean) : [];

package/src/resources/extensions/gsd/safety/safety-harness.ts

@@ -25,6 +25,8 @@ export interface SafetyHarnessConfig {
   checkpoints: boolean;
   auto_rollback: boolean;
   timeout_scale_cap: number;
+  /** Glob patterns for files excluded from unexpected-change warnings (#4385). */
+  file_change_allowlist: string[];
 }
 
 // ─── Defaults ───────────────────────────────────────────────────────────────
@@ -39,6 +41,7 @@ const DEFAULTS: SafetyHarnessConfig = {
   checkpoints: true,
   auto_rollback: false,
   timeout_scale_cap: 6,
+  file_change_allowlist: [],
 };
 
 // ─── Public API ─────────────────────────────────────────────────────────────
@@ -62,6 +65,9 @@ export function resolveSafetyHarnessConfig(
     checkpoints: typeof raw.checkpoints === "boolean" ? raw.checkpoints : DEFAULTS.checkpoints,
     auto_rollback: typeof raw.auto_rollback === "boolean" ? raw.auto_rollback : DEFAULTS.auto_rollback,
     timeout_scale_cap: typeof raw.timeout_scale_cap === "number" ? raw.timeout_scale_cap : DEFAULTS.timeout_scale_cap,
+    file_change_allowlist: Array.isArray(raw.file_change_allowlist)
+      ? (raw.file_change_allowlist as unknown[]).filter((p): p is string => typeof p === "string")
+      : DEFAULTS.file_change_allowlist,
   };
 }
 
@@ -86,6 +92,9 @@ export {
   getFilePaths,
   recordToolCall,
   recordToolResult,
+  saveEvidenceToDisk,
+  loadEvidenceFromDisk,
+  clearEvidenceFromDisk,
 } from "./evidence-collector.js";
 
 export type { EvidenceEntry, BashEvidence, FileWriteEvidence, FileEditEvidence } from "./evidence-collector.js";