gsd-pi 2.63.0-dev.351157b → 2.63.0-dev.d04bbc5

package/src/resources/extensions/gsd/safety/content-validator.ts

@@ -0,0 +1,98 @@
+/**
+ * Lightweight content validator for auto-mode safety harness.
+ * Validates that high-value unit outputs contain minimum expected content.
+ *
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+import { existsSync, readFileSync } from "node:fs";
+import { logWarning } from "../workflow-logger.js";
+
+// ─── Types ──────────────────────────────────────────────────────────────────
+
+export interface ContentViolation {
+  severity: "warning";
+  reason: string;
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/**
+ * Validate content quality for a completed unit.
+ * Returns an array of violations. Empty array = content looks acceptable.
+ *
+ * @param unitType - The type of unit that completed (e.g. "plan-slice")
+ * @param artifactPath - Absolute path to the primary artifact file
+ */
+export function validateContent(
+  unitType: string,
+  artifactPath: string | null,
+): ContentViolation[] {
+  if (!artifactPath || !existsSync(artifactPath)) return [];
+
+  const validator = VALIDATORS[unitType];
+  if (!validator) return [];
+
+  try {
+    const content = readFileSync(artifactPath, "utf-8");
+    return validator(content);
+  } catch (e) {
+    logWarning("safety", `content validation read failed: ${(e as Error).message}`);
+    return [];
+  }
+}
+
+// ─── Validators ─────────────────────────────────────────────────────────────
+
+type ContentValidatorFn = (content: string) => ContentViolation[];
+
+const VALIDATORS: Record<string, ContentValidatorFn> = {
+  "plan-slice": validatePlanSlice,
+  "plan-milestone": validatePlanMilestone,
+};
+
+function validatePlanSlice(content: string): ContentViolation[] {
+  const violations: ContentViolation[] = [];
+
+  // Must have at least 2 task entries (checkbox pattern)
+  const taskCount = (content.match(/- \[[ x]\] \*\*T\d+/g) || []).length;
+  if (taskCount < 2) {
+    violations.push({
+      severity: "warning",
+      reason: `Slice plan has only ${taskCount} task(s) — expected at least 2`,
+    });
+  }
+
+  // Should have a Files Likely Touched section
+  if (!content.includes("## Files Likely Touched") && !content.includes("## Files")) {
+    violations.push({
+      severity: "warning",
+      reason: "Slice plan missing 'Files Likely Touched' section",
+    });
+  }
+
+  // Should have a verification section
+  if (!content.includes("Verify") && !content.includes("verify")) {
+    violations.push({
+      severity: "warning",
+      reason: "Slice plan has no verification instructions",
+    });
+  }
+
+  return violations;
+}
+
+function validatePlanMilestone(content: string): ContentViolation[] {
+  const violations: ContentViolation[] = [];
+
+  // Must have at least 1 slice entry
+  const sliceCount = (content.match(/##\s+S\d+/g) || []).length;
+  if (sliceCount < 1) {
+    violations.push({
+      severity: "warning",
+      reason: `Milestone roadmap has ${sliceCount} slice(s) — expected at least 1`,
+    });
+  }
+
+  return violations;
+}

package/src/resources/extensions/gsd/safety/destructive-guard.ts

@@ -0,0 +1,49 @@
+/**
+ * Destructive command classifier for auto-mode safety harness.
+ * Classifies bash commands and warns on potentially destructive operations.
+ * Does NOT block — only classifies for logging/notification.
+ *
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+// ─── Pattern Definitions ────────────────────────────────────────────────────
+
+interface DestructivePattern {
+  pattern: RegExp;
+  label: string;
+}
+
+const DESTRUCTIVE_PATTERNS: readonly DestructivePattern[] = [
+  { pattern: /\brm\s+(-[^\s]*[rfRF][^\s]*\s+|.*\s+-[^\s]*[rfRF])/, label: "recursive delete" },
+  { pattern: /\bgit\s+push\s+.*--force/, label: "force push" },
+  { pattern: /\bgit\s+push\s+-f\b/, label: "force push" },
+  { pattern: /\bgit\s+reset\s+--hard/, label: "hard reset" },
+  { pattern: /\bgit\s+clean\s+-[^\s]*[fdxFDX]/, label: "git clean" },
+  { pattern: /\bgit\s+checkout\s+--\s+\./, label: "discard all changes" },
+  { pattern: /\bdrop\s+(database|table|index)\b/i, label: "SQL drop" },
+  { pattern: /\btruncate\s+table\b/i, label: "SQL truncate" },
+  { pattern: /\bchmod\s+777\b/, label: "world-writable permissions" },
+  { pattern: /\bcurl\s.*\|\s*(bash|sh|zsh)\b/, label: "pipe to shell" },
+];
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+export interface CommandClassification {
+  destructive: boolean;
+  labels: string[];
+}
+
+/**
+ * Classify a bash command for destructive operations.
+ * Returns the list of matched destructive pattern labels.
+ */
+export function classifyCommand(command: string): CommandClassification {
+  const labels: string[] = [];
+  for (const { pattern, label } of DESTRUCTIVE_PATTERNS) {
+    if (pattern.test(command)) {
+      // Deduplicate labels (e.g., two force-push patterns)
+      if (!labels.includes(label)) labels.push(label);
+    }
+  }
+  return { destructive: labels.length > 0, labels };
+}

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

@@ -0,0 +1,151 @@
+/**
+ * Real-time tool call evidence collector for auto-mode safety harness.
+ * 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.
+ *
+ * Follows the same module-level Map pattern as auto-tool-tracking.ts.
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+// ─── Types ──────────────────────────────────────────────────────────────────
+
+export interface BashEvidence {
+  kind: "bash";
+  toolCallId: string;
+  command: string;
+  exitCode: number;
+  outputSnippet: string;
+  timestamp: number;
+}
+
+export interface FileWriteEvidence {
+  kind: "write";
+  toolCallId: string;
+  path: string;
+  timestamp: number;
+}
+
+export interface FileEditEvidence {
+  kind: "edit";
+  toolCallId: string;
+  path: string;
+  timestamp: number;
+}
+
+export type EvidenceEntry = BashEvidence | FileWriteEvidence | FileEditEvidence;
+
+// ─── Module State ───────────────────────────────────────────────────────────
+
+let unitEvidence: EvidenceEntry[] = [];
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/** Reset all evidence for a new unit. Call at unit start. */
+export function resetEvidence(): void {
+  unitEvidence = [];
+}
+
+/** Get a read-only view of all evidence collected for the current unit. */
+export function getEvidence(): readonly EvidenceEntry[] {
+  return unitEvidence;
+}
+
+/** Get only bash evidence entries. */
+export function getBashEvidence(): readonly BashEvidence[] {
+  return unitEvidence.filter((e): e is BashEvidence => e.kind === "bash");
+}
+
+/** Get all file paths touched (write + edit). */
+export function getFilePaths(): string[] {
+  return unitEvidence
+    .filter((e): e is FileWriteEvidence | FileEditEvidence => e.kind === "write" || e.kind === "edit")
+    .map(e => e.path);
+}
+
+// ─── Recording (called from register-hooks.ts) ─────────────────────────────
+
+/**
+ * Record a tool call at dispatch time (before execution).
+ * Exit codes and output are filled in by recordToolResult after execution.
+ */
+export function recordToolCall(toolName: string, input: Record<string, unknown>): void {
+  if (toolName === "bash" || toolName === "Bash") {
+    unitEvidence.push({
+      kind: "bash",
+      toolCallId: "",
+      command: String(input.command ?? ""),
+      exitCode: -1,
+      outputSnippet: "",
+      timestamp: Date.now(),
+    });
+  } else if (toolName === "write" || toolName === "Write") {
+    unitEvidence.push({
+      kind: "write",
+      toolCallId: "",
+      path: String(input.file_path ?? input.path ?? ""),
+      timestamp: Date.now(),
+    });
+  } else if (toolName === "edit" || toolName === "Edit") {
+    unitEvidence.push({
+      kind: "edit",
+      toolCallId: "",
+      path: String(input.file_path ?? input.path ?? ""),
+      timestamp: Date.now(),
+    });
+  }
+}
+
+/**
+ * Record a tool execution result. Matches the most recent unresolved entry
+ * of the same kind and fills in the toolCallId, exit code, and output.
+ */
+export function recordToolResult(
+  toolCallId: string,
+  toolName: string,
+  result: unknown,
+  isError: boolean,
+): void {
+  const normalizedName = toolName.toLowerCase();
+
+  if (normalizedName === "bash") {
+    const entry = findLastUnresolved("bash") as BashEvidence | undefined;
+    if (entry) {
+      entry.toolCallId = toolCallId;
+      const text = extractResultText(result);
+      entry.outputSnippet = text.slice(0, 500);
+      const exitMatch = text.match(/Command exited with code (\d+)/);
+      entry.exitCode = exitMatch ? Number(exitMatch[1]) : (isError ? 1 : 0);
+    }
+  } else if (normalizedName === "write" || normalizedName === "edit") {
+    const entry = findLastUnresolved(normalizedName as "write" | "edit");
+    if (entry) {
+      entry.toolCallId = toolCallId;
+    }
+  }
+}
+
+// ─── Internals ──────────────────────────────────────────────────────────────
+
+function findLastUnresolved(kind: string): EvidenceEntry | undefined {
+  for (let i = unitEvidence.length - 1; i >= 0; i--) {
+    if (unitEvidence[i].kind === kind && unitEvidence[i].toolCallId === "") {
+      return unitEvidence[i];
+    }
+  }
+  return undefined;
+}
+
+function extractResultText(result: unknown): string {
+  if (typeof result === "string") return result;
+  if (result && typeof result === "object") {
+    const r = result as Record<string, unknown>;
+    if (Array.isArray(r.content)) {
+      const textBlock = r.content.find(
+        (c: unknown) => typeof c === "object" && c !== null && (c as Record<string, unknown>).type === "text",
+      ) as Record<string, unknown> | undefined;
+      if (textBlock && typeof textBlock.text === "string") return textBlock.text;
+    }
+    if (typeof r.text === "string") return r.text;
+  }
+  return String(result ?? "");
+}

package/src/resources/extensions/gsd/safety/evidence-cross-ref.ts

@@ -0,0 +1,120 @@
+/**
+ * Evidence cross-reference for auto-mode safety harness.
+ * Compares the LLM's claimed verification evidence (command + exitCode)
+ * against actual bash tool calls recorded by the evidence collector.
+ *
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+import type { BashEvidence, EvidenceEntry } from "./evidence-collector.js";
+
+// ─── Types ──────────────────────────────────────────────────────────────────
+
+export interface ClaimedEvidence {
+  command: string;
+  exitCode: number;
+  verdict: string;
+}
+
+export interface EvidenceMismatch {
+  severity: "warning" | "error";
+  claimed: ClaimedEvidence;
+  actual: BashEvidence | null;
+  reason: string;
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/**
+ * Cross-reference claimed verification evidence against actual bash tool calls.
+ *
+ * Returns an array of mismatches. Empty array = all claims verified.
+ * Skips entries that were coerced from strings (already flagged by db-tools.ts).
+ */
+export function crossReferenceEvidence(
+  claimedEvidence: readonly ClaimedEvidence[],
+  actualEvidence: readonly EvidenceEntry[],
+): EvidenceMismatch[] {
+  const bashCalls = actualEvidence.filter(
+    (e): e is BashEvidence => e.kind === "bash",
+  );
+  const mismatches: EvidenceMismatch[] = [];
+
+  for (const claimed of claimedEvidence) {
+    // Skip coerced entries — they're already flagged with exitCode: -1
+    // and verdict: "unknown (coerced from string)" by db-tools.ts
+    if (claimed.verdict?.includes("coerced from string")) continue;
+    if (claimed.exitCode === -1) continue;
+
+    // Skip entries with empty or generic commands
+    if (!claimed.command || claimed.command.length < 3) continue;
+
+    // Find matching bash call by command substring match
+    const match = findBestMatch(claimed.command, bashCalls);
+
+    if (!match) {
+      mismatches.push({
+        severity: "warning",
+        claimed,
+        actual: null,
+        reason: `No bash tool call found matching "${claimed.command.slice(0, 80)}"`,
+      });
+      continue;
+    }
+
+    // Exit code mismatch: LLM claims success but actual command failed
+    if (claimed.exitCode === 0 && match.exitCode !== 0) {
+      mismatches.push({
+        severity: "error",
+        claimed,
+        actual: match,
+        reason: `Claimed exitCode=0 but actual exitCode=${match.exitCode}`,
+      });
+    }
+  }
+
+  return mismatches;
+}
+
+// ─── Internals ──────────────────────────────────────────────────────────────
+
+/**
+ * Find the best matching bash evidence entry for a claimed command.
+ * Uses substring matching — the claimed command may be a shortened version
+ * of the actual command, or vice versa.
+ */
+function findBestMatch(
+  claimedCommand: string,
+  bashCalls: readonly BashEvidence[],
+): BashEvidence | null {
+  const normalized = claimedCommand.trim();
+
+  // Exact match first
+  const exact = bashCalls.find(b => b.command.trim() === normalized);
+  if (exact) return exact;
+
+  // Substring match: claimed is contained in actual or actual in claimed
+  const substring = bashCalls.find(
+    b => b.command.includes(normalized) || normalized.includes(b.command),
+  );
+  if (substring) return substring;
+
+  // Token match: split on whitespace and check significant overlap
+  const claimedTokens = normalized.split(/\s+/).filter(t => t.length > 2);
+  if (claimedTokens.length === 0) return null;
+
+  let bestMatch: BashEvidence | null = null;
+  let bestScore = 0;
+
+  for (const call of bashCalls) {
+    const callTokens = new Set(call.command.split(/\s+/));
+    const matchCount = claimedTokens.filter(t => callTokens.has(t)).length;
+    const score = matchCount / claimedTokens.length;
+    if (score > bestScore && score >= 0.5) {
+      bestScore = score;
+      bestMatch = call;
+    }
+  }
+
+  return bestMatch;
+}

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

@@ -0,0 +1,108 @@
+/**
+ * Post-unit file change validator for auto-mode safety harness.
+ * Compares actual git diff against the task plan's expected output files.
+ *
+ * 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.
+ *
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+import { execFileSync } from "node:child_process";
+import { logWarning } from "../workflow-logger.js";
+
+// ─── Types ──────────────────────────────────────────────────────────────────
+
+export interface FileViolation {
+  severity: "info" | "warning";
+  file: string;
+  reason: string;
+}
+
+export interface FileChangeAudit {
+  expectedFiles: string[];
+  actualFiles: string[];
+  unexpectedFiles: string[];
+  missingFiles: string[];
+  violations: FileViolation[];
+}
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/**
+ * Validate file changes after auto-commit for an execute-task unit.
+ * Returns null if task data is unavailable or DB is not loaded.
+ *
+ * @param basePath - Working directory (worktree or project root)
+ * @param expectedOutput - JSON array from tasks.expected_output DB column
+ * @param plannedFiles - JSON array from tasks.files DB column
+ */
+export function validateFileChanges(
+  basePath: string,
+  expectedOutput: string[],
+  plannedFiles: string[],
+): FileChangeAudit | null {
+  const allExpected = new Set([...expectedOutput, ...plannedFiles]);
+
+  // If no expected files were planned, skip validation
+  if (allExpected.size === 0) return null;
+
+  // Get actual changed files from last commit
+  const actualFiles = getChangedFilesFromLastCommit(basePath);
+  if (!actualFiles) return null;
+
+  // Filter out .gsd/ internal files — only validate project source files
+  const projectFiles = actualFiles.filter(f => !f.startsWith(".gsd/") && !f.startsWith(".gsd\\"));
+
+  // Normalize expected paths (strip leading ./ or /)
+  const normalizedExpected = new Set(
+    [...allExpected].map(f => f.replace(/^\.\//, "").replace(/^\//, "")),
+  );
+
+  // Compute symmetric difference
+  const unexpectedFiles = projectFiles.filter(f => !normalizedExpected.has(f));
+  const missingFiles = [...normalizedExpected].filter(f => !projectFiles.includes(f));
+
+  const violations: FileViolation[] = [];
+
+  for (const f of unexpectedFiles) {
+    violations.push({
+      severity: "warning",
+      file: f,
+      reason: "Modified but not in task plan's expected output",
+    });
+  }
+
+  for (const f of missingFiles) {
+    violations.push({
+      severity: "info",
+      file: f,
+      reason: "Listed in task plan but not modified",
+    });
+  }
+
+  return {
+    expectedFiles: [...normalizedExpected],
+    actualFiles: projectFiles,
+    unexpectedFiles,
+    missingFiles,
+    violations,
+  };
+}
+
+// ─── Internals ──────────────────────────────────────────────────────────────
+
+function getChangedFilesFromLastCommit(basePath: string): string[] | null {
+  try {
+    const result = execFileSync(
+      "git",
+      ["diff", "--name-only", "HEAD~1", "HEAD"],
+      { cwd: basePath, stdio: ["ignore", "pipe", "pipe"], encoding: "utf-8" },
+    ).trim();
+    return result ? result.split("\n").filter(Boolean) : [];
+  } catch (e) {
+    logWarning("safety", `git diff failed in file-change-validator: ${(e as Error).message}`);
+    return null;
+  }
+}

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

@@ -0,0 +1,106 @@
+/**
+ * Pre-unit git checkpoint and rollback for auto-mode safety harness.
+ * Uses the existing refs/gsd/ namespace (already pruned by doctor).
+ *
+ * Creates a lightweight ref at HEAD before unit execution. On failure,
+ * the ref can be used to rollback the branch to the pre-unit state.
+ *
+ * Copyright (c) 2026 Jeremy McSpadden <jeremy@fluxlabs.net>
+ */
+
+import { execFileSync } from "node:child_process";
+import { logWarning } from "../workflow-logger.js";
+
+// ─── Constants ──────────────────────────────────────────────────────────────
+
+const CHECKPOINT_PREFIX = "refs/gsd/checkpoints/";
+
+// ─── Public API ─────────────────────────────────────────────────────────────
+
+/**
+ * Create a checkpoint ref at the current HEAD for the given unit.
+ * Returns the SHA of HEAD, or null if the operation fails.
+ */
+export function createCheckpoint(basePath: string, unitId: string): string | null {
+  try {
+    const sha = execFileSync("git", ["rev-parse", "HEAD"], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+      encoding: "utf-8",
+    }).trim();
+
+    if (!sha || sha.length < 7) return null;
+
+    // Sanitize unitId for use in ref path (replace / with -)
+    const safeUnitId = unitId.replace(/\//g, "-");
+
+    execFileSync("git", ["update-ref", `${CHECKPOINT_PREFIX}${safeUnitId}`, sha], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+
+    return sha;
+  } catch (e) {
+    logWarning("safety", `checkpoint creation failed: ${(e as Error).message}`);
+    return null;
+  }
+}
+
+/**
+ * Rollback the current branch to a checkpoint SHA.
+ * Returns true on success, false on failure.
+ *
+ * WARNING: This is a destructive operation — it discards all changes
+ * since the checkpoint. Only call when the user has opted in via
+ * safety_harness.auto_rollback or an explicit manual trigger.
+ */
+export function rollbackToCheckpoint(
+  basePath: string,
+  unitId: string,
+  sha: string,
+): boolean {
+  try {
+    // Get current branch name
+    const branch = execFileSync("git", ["rev-parse", "--abbrev-ref", "HEAD"], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+      encoding: "utf-8",
+    }).trim();
+
+    if (!branch || branch === "HEAD") {
+      logWarning("safety", "rollback: detached HEAD state, cannot rollback");
+      return false;
+    }
+
+    // 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).
+    execFileSync("git", ["reset", "--hard", sha], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+
+    // Cleanup checkpoint ref
+    cleanupCheckpoint(basePath, unitId);
+
+    return true;
+  } catch (e) {
+    logWarning("safety", `rollback failed: ${(e as Error).message}`);
+    return false;
+  }
+}
+
+/**
+ * Remove a checkpoint ref after successful unit completion.
+ */
+export function cleanupCheckpoint(basePath: string, unitId: string): void {
+  try {
+    const safeUnitId = unitId.replace(/\//g, "-");
+    execFileSync("git", ["update-ref", "-d", `${CHECKPOINT_PREFIX}${safeUnitId}`], {
+      cwd: basePath,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+  } catch {
+    // Non-fatal — ref may already have been cleaned up
+  }
+}