gsd-pi 2.76.0 → 2.77.0-dev.2daa994b6

package/src/resources/extensions/gsd/tools/complete-task.ts

@@ -38,6 +38,7 @@ import { writeManifest } from "../workflow-manifest.js";
 import { appendEvent } from "../workflow-events.js";
 import { logWarning, logError } from "../workflow-logger.js";
 import { loadEffectiveGSDPreferences } from "../preferences.js";
+import { isStaleWrite } from "../auto/turn-epoch.js";
 import { buildEscalationArtifact, writeEscalationArtifact } from "../escalation.js";
 
 export interface CompleteTaskResult {
@@ -45,6 +46,14 @@ export interface CompleteTaskResult {
   sliceId: string;
   milestoneId: string;
   summaryPath: string;
+  /**
+   * True when this call re-completed an already-closed task from a turn that
+   * had been superseded by timeout recovery or cancellation. The underlying
+   * state was not mutated; the response is a no-op shaped like a success so
+   * the orphaned LLM tool call resolves cleanly.
+   */
+  duplicate?: boolean;
+  stale?: boolean;
 }
 
 import type { TaskRow } from "../gsd-db.js";
@@ -211,6 +220,18 @@ export async function handleCompleteTask(
 
     const existingTask = getTask(params.milestoneId, params.sliceId, params.taskId);
     if (existingTask && isClosedStatus(existingTask.status)) {
+      // Stale-turn path: a timed-out turn that was superseded by recovery
+      // can still reach this code when its LLM call eventually returns and
+      // invokes gsd_complete_task. Returning an error would produce noisy
+      // "already complete — use reopen first" logs in the orphaned turn.
+      // Instead, signal the duplicate via a non-mutating success shape that
+      // callers can detect via `duplicate: true` / `stale: true`.
+      if (isStaleWrite("complete-task")) {
+        // Sentinel handled below — outside the transaction — so we don't
+        // render SUMMARY.md or flip plan checkboxes for a stale duplicate.
+        guardError = "__stale_duplicate__";
+        return;
+      }
       guardError = `task ${params.taskId} is already complete — use gsd_task_reopen first if you need to redo it`;
       return;
     }
@@ -248,6 +269,34 @@ export async function handleCompleteTask(
     }
   });
 
+  if (guardError === "__stale_duplicate__") {
+    // Orphaned-turn duplicate: the task is already complete from the
+    // superseded turn's earlier (real) call. Return a non-mutating success
+    // so the stale LLM tool call unwinds cleanly. summaryPath is synthesized
+    // from the existing on-disk layout; no file is written.
+    const tasksDir = resolveTasksDir(basePath, params.milestoneId, params.sliceId);
+    const staleSummaryPath = tasksDir
+      ? join(tasksDir, `${params.taskId}-SUMMARY.md`)
+      : join(
+          basePath,
+          ".gsd",
+          "milestones",
+          params.milestoneId,
+          "slices",
+          params.sliceId,
+          "tasks",
+          `${params.taskId}-SUMMARY.md`,
+        );
+    return {
+      taskId: params.taskId,
+      sliceId: params.sliceId,
+      milestoneId: params.milestoneId,
+      summaryPath: staleSummaryPath,
+      duplicate: true,
+      stale: true,
+    };
+  }
+
   if (guardError) {
     return { error: guardError };
   }

package/src/resources/extensions/gsd/tools/exec-search-tool.ts

@@ -0,0 +1,81 @@
+// GSD Exec Search Tool — lists and filters prior gsd_exec runs.
+//
+// Scans .gsd/exec/*.meta.json and returns a ranked summary so agents can
+// re-discover past runs without re-executing. Read-only; no DB writes.
+
+import { searchExecHistory, type ExecSearchOptions } from "../exec-history.js";
+
+export interface ExecSearchToolParams {
+  query?: string;
+  runtime?: "bash" | "node" | "python";
+  failing_only?: boolean;
+  limit?: number;
+}
+
+export interface ToolExecutionResult {
+  content: Array<{ type: "text"; text: string }>;
+  details: Record<string, unknown>;
+  isError?: boolean;
+}
+
+export function executeExecSearch(
+  params: ExecSearchToolParams,
+  opts: { baseDir: string },
+): ToolExecutionResult {
+  const searchOpts: ExecSearchOptions = {
+    query: typeof params.query === "string" ? params.query : undefined,
+    runtime: params.runtime,
+    failing_only: params.failing_only === true,
+    limit: typeof params.limit === "number" ? params.limit : undefined,
+  };
+  const hits = searchExecHistory(opts.baseDir, searchOpts);
+
+  if (hits.length === 0) {
+    return {
+      content: [{ type: "text", text: "No prior gsd_exec runs match those filters." }],
+      details: { operation: "gsd_exec_search", matches: 0 },
+    };
+  }
+
+  const lines: string[] = [`Found ${hits.length} exec run(s), most recent first:`];
+  for (const hit of hits) {
+    const e = hit.entry;
+    const status = formatStatus(e);
+    const purpose = e.purpose ? ` — ${e.purpose}` : "";
+    const truncated = e.stdout_truncated ? " (stdout truncated)" : "";
+    lines.push(
+      `- [${e.id}] ${e.runtime} ${status} ${e.duration_ms}ms${truncated}${purpose}`,
+      `    stdout: ${e.stdout_path}`,
+    );
+    if (hit.digest_preview) {
+      const preview = hit.digest_preview.replace(/\n/g, "\n      ");
+      lines.push(`    preview:\n      ${preview}`);
+    }
+  }
+
+  return {
+    content: [{ type: "text", text: lines.join("\n") }],
+    details: {
+      operation: "gsd_exec_search",
+      matches: hits.length,
+      results: hits.map((hit) => ({
+        id: hit.entry.id,
+        runtime: hit.entry.runtime,
+        exit_code: hit.entry.exit_code,
+        timed_out: hit.entry.timed_out,
+        duration_ms: hit.entry.duration_ms,
+        purpose: hit.entry.purpose,
+        stdout_path: hit.entry.stdout_path,
+        stderr_path: hit.entry.stderr_path,
+        meta_path: hit.entry.meta_path,
+      })),
+    },
+  };
+}
+
+function formatStatus(entry: { exit_code: number | null; timed_out: boolean; signal: string | null }): string {
+  if (entry.timed_out) return "timeout";
+  if (entry.signal) return `signal:${entry.signal}`;
+  if (entry.exit_code === null) return "exit:null";
+  return `exit:${entry.exit_code}`;
+}

package/src/resources/extensions/gsd/tools/exec-tool.ts

@@ -0,0 +1,183 @@
+// GSD Exec Tool — executor for the gsd_exec MCP tool.
+//
+// Thin wrapper around exec-sandbox.ts that reads effective options from
+// the project preferences (context_mode block) and formats the result
+// for MCP return.
+
+import {
+  EXEC_DEFAULTS,
+  runExecSandbox,
+  type ExecSandboxOptions,
+  type ExecSandboxRequest,
+  type ExecSandboxResult,
+} from "../exec-sandbox.js";
+import { isContextModeEnabled, type ContextModeConfig } from "../preferences-types.js";
+
+export interface ExecToolParams {
+  runtime: ExecSandboxRequest["runtime"];
+  script: string;
+  purpose?: string;
+  timeout_ms?: number;
+}
+
+export interface ToolExecutionResult {
+  content: Array<{ type: "text"; text: string }>;
+  details: Record<string, unknown>;
+  isError?: boolean;
+}
+
+export interface ExecToolDeps {
+  baseDir: string;
+  preferences: { context_mode?: ContextModeConfig } | null;
+  /** Optional override for testing. */
+  run?: (req: ExecSandboxRequest, opts: ExecSandboxOptions) => Promise<ExecSandboxResult>;
+  now?: () => Date;
+  generateId?: () => string;
+}
+
+export function buildExecOptions(
+  baseDir: string,
+  cfg: ContextModeConfig | undefined,
+  extras?: Pick<ExecSandboxOptions, "env" | "now" | "generateId">,
+): ExecSandboxOptions {
+  const allowlist = Array.isArray(cfg?.exec_env_allowlist) ? cfg!.exec_env_allowlist! : EXEC_DEFAULTS.envAllowlist;
+  const stdoutCap = clampNumber(
+    cfg?.exec_stdout_cap_bytes,
+    EXEC_DEFAULTS.stdoutCapBytes,
+    4_096,
+    16_777_216,
+  );
+  const defaultTimeout = clampNumber(
+    cfg?.exec_timeout_ms,
+    EXEC_DEFAULTS.defaultTimeoutMs,
+    1_000,
+    EXEC_DEFAULTS.clampTimeoutMs,
+  );
+  const digestChars = clampNumber(cfg?.exec_digest_chars, EXEC_DEFAULTS.digestChars, 0, 4_000);
+  return {
+    baseDir,
+    clamp_timeout_ms: EXEC_DEFAULTS.clampTimeoutMs,
+    default_timeout_ms: defaultTimeout,
+    stdout_cap_bytes: stdoutCap,
+    stderr_cap_bytes: EXEC_DEFAULTS.stderrCapBytes,
+    digest_chars: digestChars,
+    env_allowlist: allowlist,
+    ...extras,
+  };
+}
+
+function clampNumber(value: unknown, fallback: number, min: number, max: number): number {
+  if (typeof value !== "number" || !Number.isFinite(value)) return fallback;
+  if (value < min) return min;
+  if (value > max) return max;
+  return Math.floor(value);
+}
+
+function isEnabled(prefs: ExecToolDeps["preferences"]): boolean {
+  return isContextModeEnabled(prefs);
+}
+
+function disabledResult(): ToolExecutionResult {
+  return {
+    content: [
+      {
+        type: "text",
+        text:
+          "gsd_exec is disabled by `context_mode.enabled: false` in preferences. Remove that " +
+          "override (or set it to true) to re-enable sandboxed tool-output execution.",
+      },
+    ],
+    details: { operation: "gsd_exec", error: "context_mode_disabled" },
+    isError: true,
+  };
+}
+
+function paramError(message: string): ToolExecutionResult {
+  return {
+    content: [{ type: "text", text: `Error: ${message}` }],
+    details: { operation: "gsd_exec", error: "invalid_params", detail: message },
+    isError: true,
+  };
+}
+
+export async function executeGsdExec(
+  params: ExecToolParams,
+  deps: ExecToolDeps,
+): Promise<ToolExecutionResult> {
+  if (!isEnabled(deps.preferences)) return disabledResult();
+
+  const runtime = params.runtime;
+  if (runtime !== "bash" && runtime !== "node" && runtime !== "python") {
+    return paramError(`invalid runtime "${String(runtime)}" — must be bash | node | python`);
+  }
+  const script = typeof params.script === "string" ? params.script : "";
+  if (script.trim().length === 0) {
+    return paramError("script is required and must be a non-empty string");
+  }
+  if (Buffer.byteLength(script, "utf8") > 200_000) {
+    return paramError("script exceeds the 200 KB length limit");
+  }
+
+  const opts = buildExecOptions(
+    deps.baseDir,
+    deps.preferences?.context_mode,
+    { now: deps.now, generateId: deps.generateId },
+  );
+  const run = deps.run ?? runExecSandbox;
+
+  try {
+    const result = await run(
+      {
+        runtime,
+        script,
+        ...(typeof params.purpose === "string" ? { purpose: params.purpose } : {}),
+        ...(typeof params.timeout_ms === "number" ? { timeout_ms: params.timeout_ms } : {}),
+      },
+      opts,
+    );
+    return formatResult(result);
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    return {
+      content: [{ type: "text", text: `Error: gsd_exec failed — ${message}` }],
+      details: { operation: "gsd_exec", error: message },
+      isError: true,
+    };
+  }
+}
+
+function formatResult(result: ExecSandboxResult): ToolExecutionResult {
+  const headerLines = [
+    `gsd_exec[${result.id}] runtime=${result.runtime} exit=${formatExit(result)} duration=${result.duration_ms}ms`,
+    `  stdout: ${result.stdout_bytes}B${result.stdout_truncated ? " (truncated)" : ""} → ${result.stdout_path}`,
+    `  stderr: ${result.stderr_bytes}B${result.stderr_truncated ? " (truncated)" : ""} → ${result.stderr_path}`,
+  ];
+  const summary = `${headerLines.join("\n")}\n--- digest ---\n${result.digest}`.trimEnd();
+  return {
+    content: [{ type: "text", text: summary }],
+    details: {
+      operation: "gsd_exec",
+      id: result.id,
+      runtime: result.runtime,
+      exit_code: result.exit_code,
+      signal: result.signal,
+      timed_out: result.timed_out,
+      duration_ms: result.duration_ms,
+      stdout_bytes: result.stdout_bytes,
+      stderr_bytes: result.stderr_bytes,
+      stdout_truncated: result.stdout_truncated,
+      stderr_truncated: result.stderr_truncated,
+      stdout_path: result.stdout_path,
+      stderr_path: result.stderr_path,
+      meta_path: result.meta_path,
+    },
+    isError: result.timed_out || result.signal !== null || result.exit_code !== 0,
+  };
+}
+
+function formatExit(result: ExecSandboxResult): string {
+  if (result.timed_out) return "timeout";
+  if (result.signal) return `signal:${result.signal}`;
+  if (result.exit_code === null) return "null";
+  return String(result.exit_code);
+}

package/src/resources/extensions/gsd/tools/memory-tools.ts

@@ -48,6 +48,14 @@ export interface MemoryCaptureParams {
   confidence?: number;
   tags?: string[];
   scope?: string;
+  /**
+   * ADR-013 Step 2: optional structured payload preserved verbatim on the
+   * memories row. Used when capturing decisions that need to retain
+   * gsd_save_decision-style fields (scope, decision, choice, rationale,
+   * made_by, revisable) so the eventual cutover (Step 6) is lossless.
+   * Plain pattern/gotcha/convention captures may omit this entirely.
+   */
+  structuredFields?: Record<string, unknown> | null;
 }
 
 const VALID_CATEGORIES = new Set([
@@ -87,7 +95,8 @@ export function executeMemoryCapture(params: MemoryCaptureParams): ToolExecution
   const scope = normalizeScope(params.scope);
   const tags = normalizeTags(params.tags);
 
-  const id = createMemory({ category, content, confidence, scope, tags });
+  const structuredFields = normalizeStructuredFields(params.structuredFields);
+  const id = createMemory({ category, content, confidence, scope, tags, structuredFields });
   if (!id) {
     return {
       content: [{ type: "text", text: "Error: failed to create memory." }],
@@ -113,6 +122,17 @@ function normalizeTags(value: unknown): string[] {
   return value.filter((t): t is string => typeof t === "string" && t.trim().length > 0).slice(0, 10);
 }
 
+function normalizeStructuredFields(value: unknown): Record<string, unknown> | null {
+  if (value == null) return null;
+  if (typeof value !== "object" || Array.isArray(value)) return null;
+  // Only accept plain objects (Object.prototype or null prototype). Class
+  // instances and exotic objects won't round-trip cleanly through JSON, so
+  // reject them here instead of producing a partially-serialized payload.
+  const proto = Object.getPrototypeOf(value);
+  if (proto !== null && proto !== Object.prototype) return null;
+  return value as Record<string, unknown>;
+}
+
 function clampConfidence(value: unknown): number {
   if (typeof value !== "number" || !Number.isFinite(value)) return 0.8;
   if (value < 0.1) return 0.1;
@@ -248,6 +268,15 @@ function includeSupersededMemories(rankedActive: Memory[]): Memory[] {
           /* leave empty */
         }
       }
+      let structuredFields: Record<string, unknown> | null = null;
+      if (typeof row["structured_fields"] === "string" && (row["structured_fields"] as string).length > 0) {
+        try {
+          const parsed = JSON.parse(row["structured_fields"] as string);
+          if (parsed && typeof parsed === "object" && !Array.isArray(parsed)) {
+            structuredFields = parsed as Record<string, unknown>;
+          }
+        } catch { /* leave null */ }
+      }
       return {
         seq: row["seq"] as number,
         id: row["id"] as string,
@@ -262,6 +291,7 @@ function includeSupersededMemories(rankedActive: Memory[]): Memory[] {
         hit_count: row["hit_count"] as number,
         scope: (row["scope"] as string) ?? "project",
         tags,
+        structured_fields: structuredFields,
       };
     });
   } catch {

package/src/resources/extensions/gsd/tools/resume-tool.ts

@@ -0,0 +1,40 @@
+// GSD Resume Tool — returns the contents of .gsd/last-snapshot.md so
+// agents can re-orient after compaction or session resume without
+// re-deriving project memory state.
+
+import { readCompactionSnapshot } from "../compaction-snapshot.js";
+
+export interface ResumeToolParams {
+  /** Ignored — reserved for future variant (e.g. dated snapshots). */
+  _variant?: string;
+}
+
+export interface ToolExecutionResult {
+  content: Array<{ type: "text"; text: string }>;
+  details: Record<string, unknown>;
+  isError?: boolean;
+}
+
+export function executeResume(
+  _params: ResumeToolParams,
+  opts: { baseDir: string },
+): ToolExecutionResult {
+  const snapshot = readCompactionSnapshot(opts.baseDir);
+  if (snapshot == null) {
+    return {
+      content: [
+        {
+          type: "text",
+          text:
+            "No snapshot found at .gsd/last-snapshot.md. The snapshot is written automatically " +
+            "on session_before_compact (enabled by default; set context_mode.enabled=false to opt out).",
+        },
+      ],
+      details: { operation: "gsd_resume", found: false },
+    };
+  }
+  return {
+    content: [{ type: "text", text: snapshot }],
+    details: { operation: "gsd_resume", found: true, bytes: Buffer.byteLength(snapshot, "utf-8") },
+  };
+}

package/src/resources/extensions/gsd/tools/validate-milestone.ts

@@ -18,6 +18,7 @@ import {
   getMilestoneSlices,
 } from "../gsd-db.js";
 import { resolveMilestonePath, clearPathCache } from "../paths.js";
+import { resolveCanonicalMilestoneRoot } from "../worktree-manager.js";
 import { saveFile, clearParseCache } from "../files.js";
 import { invalidateStateCache } from "../state.js";
 import { VALIDATION_VERDICTS, isValidMilestoneVerdict } from "../verdict-parser.js";
@@ -100,14 +101,19 @@ export async function handleValidateMilestone(
   }
 
   // ── Resolve paths and render markdown ────────────────────────────────
+  // #4761: route through the canonical-root resolver so that when a live
+  // worktree exists for this milestone, validation reads/writes the
+  // worktree's artifacts instead of stale project-root state.
   const validationMd = renderValidationMarkdown(params);
 
+  const canonicalBase = resolveCanonicalMilestoneRoot(basePath, params.milestoneId);
+
   let validationPath: string;
-  const milestoneDir = resolveMilestonePath(basePath, params.milestoneId);
+  const milestoneDir = resolveMilestonePath(canonicalBase, params.milestoneId);
   if (milestoneDir) {
     validationPath = join(milestoneDir, `${params.milestoneId}-VALIDATION.md`);
   } else {
-    const gsdDir = join(basePath, ".gsd");
+    const gsdDir = join(canonicalBase, ".gsd");
     const manualDir = join(gsdDir, "milestones", params.milestoneId);
     validationPath = join(manualDir, `${params.milestoneId}-VALIDATION.md`);
   }