Merge pull request #164 from phj1081/codex/owner/ejclaw

feat: add read-only evidence tools

runners/agent-runner/src/host-evidence.ts

@@ -4,6 +4,39 @@ import path from 'path';
 export const HOST_EVIDENCE_ACTIONS = [
   'ejclaw_service_status',
   'ejclaw_service_logs',
+  'ejclaw_deploy_state',
+  'ejclaw_artifact_metadata',
+  'db_paired_task_status',
+  'db_paired_task_flow',
+  'db_recent_paired_failures',
+  'github_pr_status',
+  'github_pr_diff_stat',
+  'github_run_status',
+] as const;
+
+export const DB_EVIDENCE_ACTIONS = [
+  'db_paired_task_status',
+  'db_paired_task_flow',
+  'db_recent_paired_failures',
+] as const;
+
+export const DEPLOY_EVIDENCE_ACTIONS = [
+  'ejclaw_deploy_state',
+  'ejclaw_artifact_metadata',
+] as const;
+
+export const GITHUB_EVIDENCE_ACTIONS = [
+  'github_pr_status',
+  'github_pr_diff_stat',
+  'github_run_status',
+] as const;
+
+export const ARTIFACT_EVIDENCE_KINDS = [
+  'build_outputs',
+  'dashboard_dist',
+  'runner_dist',
+  'android_debug_apk',
+  'attachments_dir',
 ] as const;
 
 export type HostEvidenceAction = (typeof HOST_EVIDENCE_ACTIONS)[number];

runners/agent-runner/src/ipc-host-evidence-tool.ts

@@ -0,0 +1,188 @@
+import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import { z } from 'zod';
+
+import {
+  ARTIFACT_EVIDENCE_KINDS,
+  DB_EVIDENCE_ACTIONS,
+  DEPLOY_EVIDENCE_ACTIONS,
+  formatHostEvidenceResponse,
+  GITHUB_EVIDENCE_ACTIONS,
+  HOST_EVIDENCE_ACTIONS,
+  normalizeHostEvidenceTailLines,
+  type HostEvidenceAction,
+  waitForHostEvidenceResponse,
+} from './host-evidence.js';
+
+type IpcWriter = (dir: string, data: object) => string;
+
+interface McpTextToolResult {
+  [key: string]: unknown;
+  content: Array<{ type: 'text'; text: string }>;
+  isError?: boolean;
+}
+
+interface RegisterHostEvidenceToolsOptions {
+  server: McpServer;
+  tasksDir: string;
+  responseDir: string;
+  groupFolder: string;
+  writeIpcFile: IpcWriter;
+}
+
+async function requestHostEvidence(
+  options: RegisterHostEvidenceToolsOptions,
+  payload: Record<string, unknown> & { action: HostEvidenceAction },
+): Promise<McpTextToolResult> {
+  const requestId = `host-evidence-${Date.now()}-${Math.random()
+    .toString(36)
+    .slice(2, 8)}`;
+
+  options.writeIpcFile(options.tasksDir, {
+    type: 'host_evidence_request',
+    requestId,
+    ...payload,
+    groupFolder: options.groupFolder,
+    timestamp: new Date().toISOString(),
+  });
+
+  try {
+    const response = await waitForHostEvidenceResponse(
+      options.responseDir,
+      requestId,
+    );
+    return {
+      content: [
+        {
+          type: 'text' as const,
+          text: formatHostEvidenceResponse(response),
+        },
+      ],
+      isError: !response.ok,
+    };
+  } catch (error) {
+    return {
+      content: [
+        {
+          type: 'text' as const,
+          text: error instanceof Error ? error.message : String(error),
+        },
+      ],
+      isError: true,
+    };
+  }
+}
+
+export function registerHostEvidenceTools(
+  options: RegisterHostEvidenceToolsOptions,
+): void {
+  const { server } = options;
+
+  server.tool(
+    'read_host_evidence',
+    'Read host-side deployment evidence through a narrow allowlist. Use this instead of broad shell access when reviewer/arbiter needs service status, deploy state, DB state, GitHub PR state, or artifact metadata.',
+    {
+      action: z
+        .enum(HOST_EVIDENCE_ACTIONS)
+        .describe(
+          'Allowlisted evidence action. Prefer specific read_db_evidence/read_deploy_evidence/read_github_evidence helpers when available.',
+        ),
+      tail_lines: z
+        .number()
+        .int()
+        .min(1)
+        .max(200)
+        .optional()
+        .describe('Only for ejclaw_service_logs. Defaults to 20.'),
+      task_id: z.string().optional().describe('Only for DB task actions.'),
+      minutes: z
+        .number()
+        .int()
+        .min(1)
+        .max(1440)
+        .optional()
+        .describe('Only for recent DB evidence. Defaults to 60.'),
+      limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(100)
+        .optional()
+        .describe('Only for DB evidence row limits. Defaults to 20.'),
+      repo: z
+        .string()
+        .optional()
+        .describe('Only for GitHub evidence, in owner/repo form.'),
+      pr_number: z.number().int().positive().optional(),
+      run_id: z.number().int().positive().optional(),
+      artifact_kind: z.enum(ARTIFACT_EVIDENCE_KINDS).optional(),
+    },
+    async (args) =>
+      requestHostEvidence(options, {
+        action: args.action,
+        tail_lines:
+          args.action === 'ejclaw_service_logs'
+            ? normalizeHostEvidenceTailLines(args.tail_lines)
+            : undefined,
+        task_id: args.task_id,
+        minutes: args.minutes,
+        limit: args.limit,
+        repo: args.repo,
+        pr_number: args.pr_number,
+        run_id: args.run_id,
+        artifact_kind: args.artifact_kind,
+      }),
+  );
+
+  server.tool(
+    'read_db_evidence',
+    'Read fixed DB evidence presets without arbitrary SQL. Raw message/output bodies are not returned.',
+    {
+      action: z.enum(DB_EVIDENCE_ACTIONS),
+      task_id: z
+        .string()
+        .optional()
+        .describe('Required for paired task status/flow actions.'),
+      minutes: z.number().int().min(1).max(1440).optional(),
+      limit: z.number().int().min(1).max(100).optional(),
+    },
+    async (args) =>
+      requestHostEvidence(options, {
+        action: args.action,
+        task_id: args.task_id,
+        minutes: args.minutes,
+        limit: args.limit,
+      }),
+  );
+
+  server.tool(
+    'read_deploy_evidence',
+    'Read fixed deploy/artifact evidence without shell access. Returns commit, dirty state, build artifact mtimes, sizes, and hashes where applicable.',
+    {
+      action: z.enum(DEPLOY_EVIDENCE_ACTIONS),
+      artifact_kind: z.enum(ARTIFACT_EVIDENCE_KINDS).optional(),
+    },
+    async (args) =>
+      requestHostEvidence(options, {
+        action: args.action,
+        artifact_kind: args.artifact_kind,
+      }),
+  );
+
+  server.tool(
+    'read_github_evidence',
+    'Read fixed GitHub PR/run evidence through host gh CLI without broad shell access.',
+    {
+      action: z.enum(GITHUB_EVIDENCE_ACTIONS),
+      repo: z.string().describe('GitHub repository in owner/repo form.'),
+      pr_number: z.number().int().positive().optional(),
+      run_id: z.number().int().positive().optional(),
+    },
+    async (args) =>
+      requestHostEvidence(options, {
+        action: args.action,
+        repo: args.repo,
+        pr_number: args.pr_number,
+        run_id: args.run_id,
+      }),
+  );
+}

runners/agent-runner/src/ipc-mcp-stdio.ts

@@ -15,13 +15,7 @@ import {
   DEFAULT_WATCH_CI_CONTEXT_MODE,
   normalizeWatchCiIntervalSeconds,
 } from './watch-ci.js';
-import {
-  formatHostEvidenceResponse,
-  HOST_EVIDENCE_ACTIONS,
-  normalizeHostEvidenceTailLines,
-  resolveHostEvidenceResponsesDir,
-  waitForHostEvidenceResponse,
-} from './host-evidence.js';
+import { resolveHostEvidenceResponsesDir } from './host-evidence.js';
 import {
   computeVerificationSnapshotId,
   formatVerificationResponse,
@@ -30,6 +24,8 @@ import {
 } from './verification.js';
 import { resolveIpcDirectories } from './ipc-paths.js';
 import { buildSendMessageIpcPayload } from './ipc-message.js';
+import { registerHostEvidenceTools } from './ipc-host-evidence-tool.js';
+import { registerRepoEvidenceTool } from './ipc-repo-evidence-tool.js';
 
 const { ipcDir: IPC_DIR, hostIpcDir: HOST_IPC_DIR } = resolveIpcDirectories(
   process.env,
@@ -413,69 +409,13 @@ server.tool(
   },
 );
 
-server.tool(
-  'read_host_evidence',
-  'Read host-side deployment evidence through a narrow allowlist. Use this instead of broad shell access when reviewer/arbiter needs service status or recent logs.',
-  {
-    action: z
-      .enum(HOST_EVIDENCE_ACTIONS)
-      .describe(
-        'ejclaw_service_status=systemctl --user show ejclaw, ejclaw_service_logs=recent journalctl lines',
-      ),
-    tail_lines: z
-      .number()
-      .int()
-      .min(1)
-      .max(200)
-      .optional()
-      .describe(
-        'Only for ejclaw_service_logs. Number of recent journal lines to fetch. Defaults to 20.',
-      ),
-  },
-  async (args) => {
-    const requestId = `host-evidence-${Date.now()}-${Math.random()
-      .toString(36)
-      .slice(2, 8)}`;
-
-    writeIpcFile(TASKS_DIR, {
-      type: 'host_evidence_request',
-      requestId,
-      action: args.action,
-      tail_lines:
-        args.action === 'ejclaw_service_logs'
-          ? normalizeHostEvidenceTailLines(args.tail_lines)
-          : undefined,
-      groupFolder,
-      timestamp: new Date().toISOString(),
-    });
-
-    try {
-      const response = await waitForHostEvidenceResponse(
-        HOST_EVIDENCE_RESPONSES_DIR,
-        requestId,
-      );
-      return {
-        content: [
-          {
-            type: 'text' as const,
-            text: formatHostEvidenceResponse(response),
-          },
-        ],
-        isError: !response.ok,
-      };
-    } catch (error) {
-      return {
-        content: [
-          {
-            type: 'text' as const,
-            text: error instanceof Error ? error.message : String(error),
-          },
-        ],
-        isError: true,
-      };
-    }
-  },
-);
+registerHostEvidenceTools({
+  server,
+  tasksDir: TASKS_DIR,
+  responseDir: HOST_EVIDENCE_RESPONSES_DIR,
+  groupFolder,
+  writeIpcFile,
+});
 
 server.tool(
   'run_verification',
@@ -536,6 +476,8 @@ server.tool(
   },
 );
 
+registerRepoEvidenceTool(server, REPO_ROOT);
+
 server.tool(
   'list_tasks',
   "List all scheduled tasks. From main: shows all tasks. From other groups: shows only that group's tasks.",

runners/agent-runner/src/ipc-repo-evidence-tool.ts

@@ -0,0 +1,53 @@
+import type { McpServer } from '@modelcontextprotocol/sdk/server/mcp.js';
+import { z } from 'zod';
+
+import {
+  formatRepoEvidenceResponse,
+  REPO_EVIDENCE_ACTIONS,
+  runRepoEvidenceRequestDirect,
+} from './repo-evidence.js';
+
+export function registerRepoEvidenceTool(
+  server: McpServer,
+  repoRoot: string,
+): void {
+  server.tool(
+    'read_repo_evidence',
+    'Read fixed git evidence directly from EJCLAW_WORK_DIR without broad shell access. Use this when reviewer/arbiter needs current branch, dirty state, recent commits, or a specific commit/ref.',
+    {
+      action: z
+        .enum(REPO_EVIDENCE_ACTIONS)
+        .describe(
+          'git_status=current branch and dirty state, git_head=current HEAD, git_recent_log=recent commits, git_show_ref=show a specific ref/commit',
+        ),
+      ref: z
+        .string()
+        .optional()
+        .describe('Only for git_show_ref. Defaults to HEAD.'),
+      limit: z
+        .number()
+        .int()
+        .min(1)
+        .max(30)
+        .optional()
+        .describe('Only for git_recent_log. Defaults to 10, max 30.'),
+    },
+    async (args) => {
+      const response = await runRepoEvidenceRequestDirect(repoRoot, {
+        action: args.action,
+        ref: args.ref,
+        limit: args.limit,
+      });
+
+      return {
+        content: [
+          {
+            type: 'text' as const,
+            text: formatRepoEvidenceResponse(response),
+          },
+        ],
+        isError: !response.ok,
+      };
+    },
+  );
+}

runners/agent-runner/src/repo-evidence.ts

@@ -0,0 +1,232 @@
+import { execFile } from 'child_process';
+
+export const REPO_EVIDENCE_ACTIONS = [
+  'git_status',
+  'git_head',
+  'git_recent_log',
+  'git_show_ref',
+] as const;
+
+export type RepoEvidenceAction = (typeof REPO_EVIDENCE_ACTIONS)[number];
+
+export interface RepoEvidenceRequest {
+  action: RepoEvidenceAction;
+  ref?: string;
+  limit?: number;
+}
+
+export interface RepoEvidenceResponse {
+  ok: boolean;
+  action: RepoEvidenceAction;
+  command: string;
+  stdout: string;
+  stderr: string;
+  exitCode: number;
+  workdir: string;
+  error?: string;
+}
+
+interface RepoEvidenceCommandSpec {
+  file: string;
+  args: string[];
+  commandText: string;
+}
+
+const DEFAULT_LOG_LIMIT = 10;
+const MAX_LOG_LIMIT = 30;
+const MAX_OUTPUT_CHARS = 16_000;
+const COMMAND_TIMEOUT_MS = 5_000;
+const COMMAND_MAX_BUFFER = 1024 * 1024;
+const SAFE_GIT_REF_PATTERN = /^[A-Za-z0-9._/@:{}~^+-]{1,160}$/;
+
+export function normalizeRepoEvidenceLimit(value?: number): number {
+  if (!Number.isFinite(value)) {
+    return DEFAULT_LOG_LIMIT;
+  }
+  const normalized = Math.trunc(value as number);
+  return Math.min(Math.max(normalized, 1), MAX_LOG_LIMIT);
+}
+
+export function normalizeRepoEvidenceRef(value?: string): string {
+  const ref = value?.trim() || 'HEAD';
+  if (
+    ref.startsWith('-') ||
+    ref.includes('..') ||
+    !SAFE_GIT_REF_PATTERN.test(ref)
+  ) {
+    throw new Error(`Unsupported git ref for read-only evidence: ${ref}`);
+  }
+  return ref;
+}
+
+export function buildRepoEvidenceCommand(
+  repoRoot: string,
+  request: RepoEvidenceRequest,
+): RepoEvidenceCommandSpec {
+  const gitArgs = ['-C', repoRoot];
+  switch (request.action) {
+    case 'git_status': {
+      const args = [...gitArgs, 'status', '--short', '--branch'];
+      return {
+        file: 'git',
+        args,
+        commandText: `git ${args.join(' ')}`,
+      };
+    }
+    case 'git_head': {
+      const args = [...gitArgs, 'log', '-1', '--oneline', '--decorate'];
+      return {
+        file: 'git',
+        args,
+        commandText: `git ${args.join(' ')}`,
+      };
+    }
+    case 'git_recent_log': {
+      const limit = normalizeRepoEvidenceLimit(request.limit);
+      const args = [...gitArgs, 'log', `-${limit}`, '--oneline', '--decorate'];
+      return {
+        file: 'git',
+        args,
+        commandText: `git ${args.join(' ')}`,
+      };
+    }
+    case 'git_show_ref': {
+      const ref = normalizeRepoEvidenceRef(request.ref);
+      const args = [
+        ...gitArgs,
+        'show',
+        '--stat',
+        '--oneline',
+        '--decorate',
+        '--no-ext-diff',
+        '--no-renames',
+        ref,
+      ];
+      return {
+        file: 'git',
+        args,
+        commandText: `git ${args.join(' ')}`,
+      };
+    }
+  }
+}
+
+function truncateRepoEvidenceText(value: string | undefined): string {
+  if (!value) return '';
+  if (value.length <= MAX_OUTPUT_CHARS) {
+    return value;
+  }
+  return `${value.slice(0, MAX_OUTPUT_CHARS)}\n...[truncated]`;
+}
+
+function execFileCapture(
+  file: string,
+  args: string[],
+): Promise<{ stdout: string; stderr: string }> {
+  return new Promise((resolve, reject) => {
+    execFile(
+      file,
+      args,
+      {
+        encoding: 'utf8',
+        timeout: COMMAND_TIMEOUT_MS,
+        maxBuffer: COMMAND_MAX_BUFFER,
+      },
+      (error, stdout, stderr) => {
+        if (error) {
+          reject(
+            Object.assign(error, {
+              stdout,
+              stderr,
+            }),
+          );
+          return;
+        }
+        resolve({ stdout, stderr });
+      },
+    );
+  });
+}
+
+function extractExitCode(error: unknown): number {
+  if (typeof error === 'object' && error !== null && 'code' in error) {
+    const code = (error as { code?: unknown }).code;
+    if (typeof code === 'number') {
+      return code;
+    }
+  }
+  return 1;
+}
+
+export async function runRepoEvidenceRequestDirect(
+  repoRoot: string,
+  request: RepoEvidenceRequest,
+): Promise<RepoEvidenceResponse> {
+  let command: RepoEvidenceCommandSpec;
+  try {
+    command = buildRepoEvidenceCommand(repoRoot, request);
+  } catch (error) {
+    return {
+      ok: false,
+      action: request.action,
+      command: '',
+      stdout: '',
+      stderr: '',
+      exitCode: 1,
+      workdir: repoRoot,
+      error: error instanceof Error ? error.message : String(error),
+    };
+  }
+
+  try {
+    const { stdout, stderr } = await execFileCapture(
+      command.file,
+      command.args,
+    );
+    return {
+      ok: true,
+      action: request.action,
+      command: command.commandText,
+      stdout: truncateRepoEvidenceText(stdout),
+      stderr: truncateRepoEvidenceText(stderr),
+      exitCode: 0,
+      workdir: repoRoot,
+    };
+  } catch (error) {
+    const stdout =
+      typeof error === 'object' && error !== null && 'stdout' in error
+        ? String((error as { stdout?: unknown }).stdout ?? '')
+        : '';
+    const stderr =
+      typeof error === 'object' && error !== null && 'stderr' in error
+        ? String((error as { stderr?: unknown }).stderr ?? '')
+        : '';
+
+    return {
+      ok: false,
+      action: request.action,
+      command: command.commandText,
+      stdout: truncateRepoEvidenceText(stdout),
+      stderr: truncateRepoEvidenceText(stderr),
+      exitCode: extractExitCode(error),
+      workdir: repoRoot,
+      error: error instanceof Error ? error.message : String(error),
+    };
+  }
+}
+
+export function formatRepoEvidenceResponse(
+  response: RepoEvidenceResponse,
+): string {
+  const parts = [
+    `Repo evidence action: ${response.action}`,
+    `Workdir: ${response.workdir}`,
+    `Exit code: ${response.exitCode}`,
+    response.command ? `$ ${response.command}` : null,
+    response.stdout ? response.stdout.trimEnd() : null,
+    response.stderr ? `[stderr]\n${response.stderr.trimEnd()}` : null,
+    response.error ? `[error] ${response.error}` : null,
+  ].filter((value): value is string => Boolean(value));
+
+  return parts.join('\n');
+}

runners/agent-runner/test/repo-evidence.test.ts

@@ -0,0 +1,144 @@
+import { execFileSync } from 'child_process';
+import fs from 'fs';
+import os from 'os';
+import path from 'path';
+
+import { afterEach, beforeEach, describe, expect, it } from 'vitest';
+
+import {
+  buildRepoEvidenceCommand,
+  formatRepoEvidenceResponse,
+  normalizeRepoEvidenceLimit,
+  normalizeRepoEvidenceRef,
+  runRepoEvidenceRequestDirect,
+} from '../src/repo-evidence.js';
+
+describe('repo evidence helpers', () => {
+  let repoDir: string;
+
+  beforeEach(() => {
+    repoDir = fs.mkdtempSync(path.join(os.tmpdir(), 'ejclaw-repo-evidence-'));
+    execFileSync('git', ['init'], { cwd: repoDir, stdio: 'ignore' });
+    execFileSync('git', ['config', 'user.email', 'test@example.com'], {
+      cwd: repoDir,
+      stdio: 'ignore',
+    });
+    execFileSync('git', ['config', 'user.name', 'Test User'], {
+      cwd: repoDir,
+      stdio: 'ignore',
+    });
+    fs.writeFileSync(path.join(repoDir, 'README.md'), 'initial\n');
+    execFileSync('git', ['add', 'README.md'], {
+      cwd: repoDir,
+      stdio: 'ignore',
+    });
+    execFileSync('git', ['commit', '-m', 'initial commit'], {
+      cwd: repoDir,
+      stdio: 'ignore',
+    });
+  });
+
+  afterEach(() => {
+    fs.rmSync(repoDir, { recursive: true, force: true });
+  });
+
+  it('normalizes bounded log limits', () => {
+    expect(normalizeRepoEvidenceLimit()).toBe(10);
+    expect(normalizeRepoEvidenceLimit(0)).toBe(1);
+    expect(normalizeRepoEvidenceLimit(3.8)).toBe(3);
+    expect(normalizeRepoEvidenceLimit(100)).toBe(30);
+  });
+
+  it('rejects refs that can change command behavior or request ranges', () => {
+    expect(normalizeRepoEvidenceRef()).toBe('HEAD');
+    expect(normalizeRepoEvidenceRef(' HEAD~1 ')).toBe('HEAD~1');
+    expect(() => normalizeRepoEvidenceRef('--help')).toThrow(
+      'Unsupported git ref',
+    );
+    expect(() => normalizeRepoEvidenceRef('main..HEAD')).toThrow(
+      'Unsupported git ref',
+    );
+    expect(() => normalizeRepoEvidenceRef('HEAD;rm')).toThrow(
+      'Unsupported git ref',
+    );
+  });
+
+  it('builds fixed git commands without invoking a shell', () => {
+    expect(buildRepoEvidenceCommand(repoDir, { action: 'git_status' })).toEqual(
+      {
+        file: 'git',
+        args: ['-C', repoDir, 'status', '--short', '--branch'],
+        commandText: `git -C ${repoDir} status --short --branch`,
+      },
+    );
+
+    expect(
+      buildRepoEvidenceCommand(repoDir, {
+        action: 'git_recent_log',
+        limit: 100,
+      }).args,
+    ).toContain('-30');
+
+    expect(
+      buildRepoEvidenceCommand(repoDir, {
+        action: 'git_show_ref',
+        ref: 'HEAD',
+      }).args,
+    ).toEqual([
+      '-C',
+      repoDir,
+      'show',
+      '--stat',
+      '--oneline',
+      '--decorate',
+      '--no-ext-diff',
+      '--no-renames',
+      'HEAD',
+    ]);
+  });
+
+  it('reads read-only git evidence from the target repository', async () => {
+    fs.appendFileSync(path.join(repoDir, 'README.md'), 'dirty\n');
+
+    const status = await runRepoEvidenceRequestDirect(repoDir, {
+      action: 'git_status',
+    });
+    expect(status.ok).toBe(true);
+    expect(status.stdout).toContain('README.md');
+
+    const head = await runRepoEvidenceRequestDirect(repoDir, {
+      action: 'git_head',
+    });
+    expect(head.ok).toBe(true);
+    expect(head.stdout).toContain('initial commit');
+
+    const log = await runRepoEvidenceRequestDirect(repoDir, {
+      action: 'git_recent_log',
+      limit: 2,
+    });
+    expect(log.ok).toBe(true);
+    expect(log.stdout).toContain('initial commit');
+
+    const shown = await runRepoEvidenceRequestDirect(repoDir, {
+      action: 'git_show_ref',
+      ref: 'HEAD',
+    });
+    expect(shown.ok).toBe(true);
+    expect(shown.stdout).toContain('initial commit');
+    expect(shown.stdout).toContain('README.md');
+  });
+
+  it('formats failures into MCP-friendly evidence', async () => {
+    const response = await runRepoEvidenceRequestDirect(repoDir, {
+      action: 'git_show_ref',
+      ref: '--help',
+    });
+
+    expect(response.ok).toBe(false);
+    const text = formatRepoEvidenceResponse(response);
+    expect(text).toContain('Repo evidence action: git_show_ref');
+    expect(text).toContain(`Workdir: ${repoDir}`);
+    expect(text).toContain('Exit code: 1');
+    expect(text).toContain('[error] Unsupported git ref');
+  });
+});

src/db-evidence.test.ts

@@ -0,0 +1,177 @@
+import { describe, expect, it } from 'vitest';
+
+import { _initTestDatabase, insertPairedTurnOutput } from './db.js';
+import { requireDatabase } from './db/runtime-database.js';
+import {
+  normalizeDbEvidenceLimit,
+  normalizeDbEvidenceMinutes,
+  normalizeDbEvidenceTaskId,
+  runDbEvidenceRequest,
+} from './db-evidence.js';
+
+describe('DB evidence presets', () => {
+  function seedPairedTask(): void {
+    _initTestDatabase();
+    requireDatabase()
+      .prepare(
+        `
+          INSERT INTO paired_tasks (
+            id, chat_jid, group_folder, owner_service_id, reviewer_service_id,
+            owner_agent_type, reviewer_agent_type, arbiter_agent_type,
+            status, created_at, updated_at
+          )
+          VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        `,
+      )
+      .run(
+        'task-1',
+        'room-1',
+        'room-folder',
+        'codex-main',
+        'claude',
+        'codex',
+        'claude-code',
+        'codex',
+        'active',
+        '2026-05-26T00:00:00.000Z',
+        '2026-05-26T00:10:00.000Z',
+      );
+    requireDatabase()
+      .prepare(
+        `
+          INSERT INTO paired_turns (
+            turn_id, task_id, task_updated_at, role, intent_kind, created_at, updated_at
+          )
+          VALUES (?, ?, ?, ?, ?, ?, ?)
+        `,
+      )
+      .run(
+        'turn-1',
+        'task-1',
+        '2026-05-26T00:10:00.000Z',
+        'owner',
+        'owner-turn',
+        '2026-05-26T00:00:01.000Z',
+        '2026-05-26T00:00:02.000Z',
+      );
+    requireDatabase()
+      .prepare(
+        `
+          INSERT INTO paired_turn_attempts (
+            attempt_id, turn_id, attempt_no, task_id, task_updated_at, role, intent_kind,
+            state, executor_service_id, executor_agent_type, active_run_id,
+            created_at, updated_at, completed_at, last_error
+          )
+          VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        `,
+      )
+      .run(
+        'turn-1:attempt:1',
+        'turn-1',
+        1,
+        'task-1',
+        '2026-05-26T00:10:00.000Z',
+        'owner',
+        'owner-turn',
+        'failed',
+        'codex-main',
+        'codex',
+        null,
+        '2026-05-26T00:00:01.000Z',
+        '2026-05-26T00:00:03.000Z',
+        '2026-05-26T00:00:04.000Z',
+        'failure with sk-12345678901234567890',
+      );
+    insertPairedTurnOutput(
+      'task-1',
+      1,
+      'owner',
+      'SECRET USER TEXT SHOULD NOT BE RETURNED',
+      '2026-05-26T00:00:05.000Z',
+    );
+    requireDatabase()
+      .prepare(
+        `
+          INSERT INTO work_items (
+            group_folder, chat_jid, agent_type, service_id, delivery_role,
+            status, start_seq, end_seq, result_payload, delivery_attempts,
+            created_at, updated_at
+          )
+          VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)
+        `,
+      )
+      .run(
+        'room-folder',
+        'room-1',
+        'codex',
+        'codex-main',
+        'owner',
+        'produced',
+        null,
+        null,
+        'raw delivery body should not be returned',
+        0,
+        '2026-05-26T00:00:06.000Z',
+        '2026-05-26T00:00:06.000Z',
+      );
+  }
+
+  it('normalizes request bounds and validates task ids', () => {
+    expect(normalizeDbEvidenceMinutes()).toBe(60);
+    expect(normalizeDbEvidenceMinutes(0)).toBe(1);
+    expect(normalizeDbEvidenceMinutes(99999)).toBe(1440);
+    expect(normalizeDbEvidenceLimit()).toBe(20);
+    expect(normalizeDbEvidenceLimit(999)).toBe(100);
+    expect(normalizeDbEvidenceTaskId(' task-1 ')).toBe('task-1');
+    expect(() => normalizeDbEvidenceTaskId('../bad task')).toThrow(
+      'Unsupported paired task id',
+    );
+  });
+
+  it('returns task status and flow metadata without raw bodies', () => {
+    seedPairedTask();
+
+    const status = JSON.parse(
+      runDbEvidenceRequest(
+        requireDatabase(),
+        { action: 'db_paired_task_status', taskId: 'task-1' },
+        { sourceGroup: 'room-folder', isMain: false },
+      ),
+    ) as { task: { id: string; group_folder: string } };
+    expect(status.task.id).toBe('task-1');
+    expect(status.task.group_folder).toBe('room-folder');
+
+    const flowText = runDbEvidenceRequest(
+      requireDatabase(),
+      { action: 'db_paired_task_flow', taskId: 'task-1' },
+      { sourceGroup: 'room-folder', isMain: false },
+    );
+    expect(flowText).toContain('"output_chars"');
+    expect(flowText).toContain('"last_error_chars"');
+    expect(flowText).not.toContain('SECRET USER TEXT');
+    expect(flowText).not.toContain('raw delivery body');
+    expect(flowText).not.toContain('sk-12345678901234567890');
+  });
+
+  it('scopes non-main DB evidence to the source group', () => {
+    seedPairedTask();
+
+    const blocked = JSON.parse(
+      runDbEvidenceRequest(
+        requireDatabase(),
+        { action: 'db_paired_task_status', taskId: 'task-1' },
+        { sourceGroup: 'other-folder', isMain: false },
+      ),
+    ) as { task: unknown };
+    expect(blocked.task).toBeNull();
+
+    const main = JSON.parse(
+      runDbEvidenceRequest(
+        requireDatabase(),
+        { action: 'db_paired_task_status', taskId: 'task-1' },
+        { sourceGroup: 'other-folder', isMain: true },
+      ),
+    ) as { task: { id: string } };
+    expect(main.task.id).toBe('task-1');
+  });
+});

src/db-evidence.ts

@@ -0,0 +1,347 @@
+import type { Database } from 'bun:sqlite';
+
+type SqlBinding = string | number | bigint | boolean | null | Uint8Array;
+
+export const DB_EVIDENCE_ACTIONS = [
+  'db_paired_task_status',
+  'db_paired_task_flow',
+  'db_recent_paired_failures',
+] as const;
+
+export type DbEvidenceAction = (typeof DB_EVIDENCE_ACTIONS)[number];
+
+export interface DbEvidenceRequest {
+  action: DbEvidenceAction;
+  taskId?: string;
+  minutes?: number;
+  limit?: number;
+}
+
+export interface DbEvidenceScope {
+  sourceGroup: string;
+  isMain: boolean;
+}
+
+const DEFAULT_RECENT_MINUTES = 60;
+const MAX_RECENT_MINUTES = 24 * 60;
+const DEFAULT_ROW_LIMIT = 20;
+const MAX_ROW_LIMIT = 100;
+const TASK_ID_PATTERN = /^[A-Za-z0-9._:@/-]{1,200}$/;
+
+export function isDbEvidenceAction(value: unknown): value is DbEvidenceAction {
+  return (
+    typeof value === 'string' &&
+    DB_EVIDENCE_ACTIONS.includes(value as DbEvidenceAction)
+  );
+}
+
+export function normalizeDbEvidenceMinutes(value?: number): number {
+  if (!Number.isFinite(value)) {
+    return DEFAULT_RECENT_MINUTES;
+  }
+  const normalized = Math.trunc(value as number);
+  return Math.min(Math.max(normalized, 1), MAX_RECENT_MINUTES);
+}
+
+export function normalizeDbEvidenceLimit(value?: number): number {
+  if (!Number.isFinite(value)) {
+    return DEFAULT_ROW_LIMIT;
+  }
+  const normalized = Math.trunc(value as number);
+  return Math.min(Math.max(normalized, 1), MAX_ROW_LIMIT);
+}
+
+export function normalizeDbEvidenceTaskId(value?: string): string {
+  const taskId = value?.trim();
+  if (!taskId || !TASK_ID_PATTERN.test(taskId)) {
+    throw new Error(`Unsupported paired task id for DB evidence: ${value}`);
+  }
+  return taskId;
+}
+
+function stringifyEvidence(value: unknown): string {
+  return JSON.stringify(value, null, 2);
+}
+
+function groupScopeClause(scope: DbEvidenceScope): {
+  clause: string;
+  params: SqlBinding[];
+} {
+  return scope.isMain
+    ? { clause: '', params: [] }
+    : { clause: ' AND group_folder = ?', params: [scope.sourceGroup] };
+}
+
+function getScopedTask(
+  database: Database,
+  taskId: string,
+  scope: DbEvidenceScope,
+): Record<string, unknown> | null {
+  const groupScope = groupScopeClause(scope);
+  const row = database
+    .prepare(
+      `
+          SELECT id,
+                 group_folder,
+                 chat_jid,
+                 status,
+                 round_trip_count,
+                 owner_failure_count,
+                 owner_step_done_streak,
+                 finalize_step_done_count,
+                 task_done_then_user_reopen_count,
+                 empty_step_done_streak,
+                 arbiter_verdict,
+                 arbiter_requested_at,
+                 completion_reason,
+                 owner_agent_type,
+                 reviewer_agent_type,
+                 arbiter_agent_type,
+                 created_at,
+                 updated_at
+            FROM paired_tasks
+           WHERE id = ?
+                 ${groupScope.clause}
+        `,
+    )
+    .get(taskId, ...groupScope.params) as Record<string, unknown> | undefined;
+  return row ?? null;
+}
+
+function runPairedTaskStatus(
+  database: Database,
+  request: DbEvidenceRequest,
+  scope: DbEvidenceScope,
+): string {
+  const taskId = normalizeDbEvidenceTaskId(request.taskId);
+  const task = getScopedTask(database, taskId, scope);
+  return stringifyEvidence({
+    action: request.action,
+    task,
+  });
+}
+
+function runPairedTaskFlow(
+  database: Database,
+  request: DbEvidenceRequest,
+  scope: DbEvidenceScope,
+): string {
+  const taskId = normalizeDbEvidenceTaskId(request.taskId);
+  const limit = normalizeDbEvidenceLimit(request.limit);
+  const task = getScopedTask(database, taskId, scope);
+  if (!task) {
+    return stringifyEvidence({
+      action: request.action,
+      task: null,
+      turns: [],
+      attempts: [],
+      outputs: [],
+      deliveries: [],
+    });
+  }
+
+  const turns = database
+    .prepare(
+      `
+        SELECT turn_id,
+               role,
+               intent_kind,
+               created_at,
+               updated_at
+          FROM paired_turns
+         WHERE task_id = ?
+         ORDER BY updated_at, created_at, turn_id
+         LIMIT ?
+      `,
+    )
+    .all(taskId, limit);
+
+  const attempts = database
+    .prepare(
+      `
+        SELECT attempt_id,
+               parent_attempt_id,
+               turn_id,
+               attempt_no,
+               role,
+               intent_kind,
+               state,
+               executor_agent_type,
+               active_run_id,
+               created_at,
+               updated_at,
+               completed_at,
+               length(last_error) AS last_error_chars
+          FROM paired_turn_attempts
+         WHERE task_id = ?
+         ORDER BY updated_at, attempt_no, attempt_id
+         LIMIT ?
+      `,
+    )
+    .all(taskId, limit);
+
+  const outputs = database
+    .prepare(
+      `
+        SELECT turn_number,
+               role,
+               verdict,
+               created_at,
+               length(output_text) AS output_chars
+          FROM paired_turn_outputs
+         WHERE task_id = ?
+         ORDER BY turn_number, created_at, role
+         LIMIT ?
+      `,
+    )
+    .all(taskId, limit);
+
+  const deliveries = database
+    .prepare(
+      `
+        SELECT id,
+               agent_type,
+               delivery_role,
+               status,
+               delivery_attempts,
+               created_at,
+               updated_at,
+               delivered_at,
+               length(result_payload) AS result_payload_chars,
+               length(last_error) AS last_error_chars
+          FROM work_items
+         WHERE chat_jid = ?
+           AND created_at >= ?
+           AND created_at <= datetime(?, '+1 minute')
+         ORDER BY created_at, id
+         LIMIT ?
+      `,
+    )
+    .all(
+      String(task.chat_jid),
+      String(task.created_at),
+      String(task.updated_at),
+      limit,
+    );
+
+  return stringifyEvidence({
+    action: request.action,
+    task,
+    turns,
+    attempts,
+    outputs,
+    deliveries,
+  });
+}
+
+function runRecentPairedFailures(
+  database: Database,
+  request: DbEvidenceRequest,
+  scope: DbEvidenceScope,
+): string {
+  const minutes = normalizeDbEvidenceMinutes(request.minutes);
+  const limit = normalizeDbEvidenceLimit(request.limit);
+  const cutoff = new Date(Date.now() - minutes * 60_000).toISOString();
+  const groupScope = groupScopeClause(scope);
+
+  const tasks = database
+    .prepare(
+      `
+        SELECT id,
+               group_folder,
+               chat_jid,
+               status,
+               owner_failure_count,
+               arbiter_verdict,
+               completion_reason,
+               arbiter_requested_at,
+               updated_at
+          FROM paired_tasks
+         WHERE updated_at >= ?
+           AND (
+             owner_failure_count > 0
+             OR status = 'arbiter_requested'
+             OR arbiter_verdict IS NOT NULL
+             OR completion_reason IS NOT NULL
+           )
+           ${groupScope.clause}
+         ORDER BY updated_at DESC
+         LIMIT ?
+      `,
+    )
+    .all(cutoff, ...groupScope.params, limit);
+
+  const attempts = database
+    .prepare(
+      `
+        SELECT attempts.task_id,
+               attempts.turn_id,
+               attempts.attempt_no,
+               attempts.role,
+               attempts.intent_kind,
+               attempts.state,
+               attempts.executor_agent_type,
+               attempts.active_run_id,
+               attempts.updated_at,
+               attempts.completed_at,
+               length(attempts.last_error) AS last_error_chars
+          FROM paired_turn_attempts attempts
+          JOIN paired_tasks tasks
+            ON tasks.id = attempts.task_id
+         WHERE attempts.updated_at >= ?
+           AND (
+             attempts.state = 'failed'
+             OR attempts.last_error IS NOT NULL
+           )
+           ${scope.isMain ? '' : 'AND tasks.group_folder = ?'}
+         ORDER BY attempts.updated_at DESC
+         LIMIT ?
+      `,
+    )
+    .all(cutoff, ...(scope.isMain ? [] : [scope.sourceGroup]), limit);
+
+  const deliveryRetries = database
+    .prepare(
+      `
+        SELECT id,
+               chat_jid,
+               agent_type,
+               delivery_role,
+               status,
+               delivery_attempts,
+               updated_at,
+               length(last_error) AS last_error_chars
+          FROM work_items
+         WHERE updated_at >= ?
+           AND status = 'delivery_retry'
+           ${scope.isMain ? '' : 'AND group_folder = ?'}
+         ORDER BY updated_at DESC
+         LIMIT ?
+      `,
+    )
+    .all(cutoff, ...(scope.isMain ? [] : [scope.sourceGroup]), limit);
+
+  return stringifyEvidence({
+    action: request.action,
+    window_minutes: minutes,
+    cutoff,
+    tasks,
+    attempts,
+    delivery_retries: deliveryRetries,
+  });
+}
+
+export function runDbEvidenceRequest(
+  database: Database,
+  request: DbEvidenceRequest,
+  scope: DbEvidenceScope,
+): string {
+  switch (request.action) {
+    case 'db_paired_task_status':
+      return runPairedTaskStatus(database, request, scope);
+    case 'db_paired_task_flow':
+      return runPairedTaskFlow(database, request, scope);
+    case 'db_recent_paired_failures':
+      return runRecentPairedFailures(database, request, scope);
+  }
+}

src/deploy-evidence.test.ts

@@ -0,0 +1,45 @@
+import fs from 'fs';
+import os from 'os';
+import path from 'path';
+
+import { describe, expect, it } from 'vitest';
+
+import {
+  collectArtifactMetadata,
+  normalizeArtifactEvidenceKind,
+} from './deploy-evidence.js';
+
+describe('deploy evidence helpers', () => {
+  it('normalizes fixed artifact kinds', () => {
+    expect(normalizeArtifactEvidenceKind()).toBe('build_outputs');
+    expect(normalizeArtifactEvidenceKind('dashboard_dist')).toBe(
+      'dashboard_dist',
+    );
+    expect(() => normalizeArtifactEvidenceKind('/etc/passwd')).toThrow(
+      'Unsupported artifact evidence kind',
+    );
+  });
+
+  it('returns metadata only for build artifacts', () => {
+    const root = fs.mkdtempSync(path.join(os.tmpdir(), 'ejclaw-deploy-'));
+    const dashboardDir = path.join(root, 'apps', 'dashboard', 'dist');
+    fs.mkdirSync(path.join(dashboardDir, 'assets'), { recursive: true });
+    fs.writeFileSync(path.join(dashboardDir, 'index.html'), '<html></html>');
+    fs.writeFileSync(path.join(dashboardDir, 'assets', 'index.js'), 'secret');
+
+    const text = collectArtifactMetadata(
+      {
+        projectRoot: root,
+        dataDir: path.join(root, 'data'),
+        dashboardStaticDir: dashboardDir,
+      },
+      { action: 'ejclaw_artifact_metadata', artifactKind: 'dashboard_dist' },
+    );
+
+    expect(text).toContain('"file_count"');
+    expect(text).toContain('"total_bytes"');
+    expect(text).toContain('index.js');
+    expect(text).not.toContain('"secret"');
+    fs.rmSync(root, { recursive: true, force: true });
+  });
+});

src/deploy-evidence.ts

@@ -0,0 +1,271 @@
+import { createHash } from 'crypto';
+import { execFile } from 'child_process';
+import fs from 'fs';
+import path from 'path';
+
+export const DEPLOY_EVIDENCE_ACTIONS = [
+  'ejclaw_deploy_state',
+  'ejclaw_artifact_metadata',
+] as const;
+
+export const ARTIFACT_EVIDENCE_KINDS = [
+  'build_outputs',
+  'dashboard_dist',
+  'runner_dist',
+  'android_debug_apk',
+  'attachments_dir',
+] as const;
+
+export type DeployEvidenceAction = (typeof DEPLOY_EVIDENCE_ACTIONS)[number];
+export type ArtifactEvidenceKind = (typeof ARTIFACT_EVIDENCE_KINDS)[number];
+
+export interface DeployEvidenceRequest {
+  action: DeployEvidenceAction;
+  artifactKind?: string;
+}
+
+export interface DeployEvidencePaths {
+  projectRoot: string;
+  dataDir: string;
+  dashboardStaticDir: string;
+}
+
+const COMMAND_TIMEOUT_MS = 5_000;
+const COMMAND_MAX_BUFFER = 1024 * 1024;
+const MAX_DIR_ENTRIES = 5_000;
+const MAX_LATEST_FILES = 12;
+
+export function isDeployEvidenceAction(
+  value: unknown,
+): value is DeployEvidenceAction {
+  return (
+    typeof value === 'string' &&
+    DEPLOY_EVIDENCE_ACTIONS.includes(value as DeployEvidenceAction)
+  );
+}
+
+export function normalizeArtifactEvidenceKind(
+  value?: string,
+): ArtifactEvidenceKind {
+  if (!value) return 'build_outputs';
+  if (ARTIFACT_EVIDENCE_KINDS.includes(value as ArtifactEvidenceKind)) {
+    return value as ArtifactEvidenceKind;
+  }
+  throw new Error(`Unsupported artifact evidence kind: ${value}`);
+}
+
+function execFileText(file: string, args: string[]): Promise<string> {
+  return new Promise((resolve, reject) => {
+    execFile(
+      file,
+      args,
+      {
+        encoding: 'utf8',
+        timeout: COMMAND_TIMEOUT_MS,
+        maxBuffer: COMMAND_MAX_BUFFER,
+      },
+      (error, stdout, stderr) => {
+        if (error) {
+          const details = stderr?.trim() || stdout?.trim() || error.message;
+          reject(new Error(`${file} ${args.join(' ')} failed: ${details}`));
+          return;
+        }
+        resolve(stdout.trim());
+      },
+    );
+  });
+}
+
+function sha256File(filePath: string): string {
+  const hash = createHash('sha256');
+  hash.update(fs.readFileSync(filePath));
+  return hash.digest('hex');
+}
+
+function fileMetadata(filePath: string): Record<string, unknown> {
+  if (!fs.existsSync(filePath)) {
+    return {
+      path: filePath,
+      exists: false,
+    };
+  }
+
+  const stat = fs.statSync(filePath);
+  const base = {
+    path: filePath,
+    exists: true,
+    type: stat.isDirectory() ? 'directory' : stat.isFile() ? 'file' : 'other',
+    size_bytes: stat.size,
+    mtime: stat.mtime.toISOString(),
+  };
+
+  if (!stat.isFile()) {
+    return base;
+  }
+
+  return {
+    ...base,
+    sha256: sha256File(filePath),
+  };
+}
+
+function walkDirectory(root: string): Array<{ path: string; stat: fs.Stats }> {
+  if (!fs.existsSync(root)) return [];
+  const pending = [root];
+  const files: Array<{ path: string; stat: fs.Stats }> = [];
+
+  while (pending.length > 0 && files.length < MAX_DIR_ENTRIES) {
+    const current = pending.pop()!;
+    for (const entry of fs.readdirSync(current, { withFileTypes: true })) {
+      const entryPath = path.join(current, entry.name);
+      if (entry.isDirectory()) {
+        pending.push(entryPath);
+        continue;
+      }
+      if (!entry.isFile()) continue;
+      files.push({ path: entryPath, stat: fs.statSync(entryPath) });
+      if (files.length >= MAX_DIR_ENTRIES) break;
+    }
+  }
+
+  return files;
+}
+
+function directoryMetadata(root: string): Record<string, unknown> {
+  if (!fs.existsSync(root)) {
+    return {
+      path: root,
+      exists: false,
+    };
+  }
+  const files = walkDirectory(root);
+  const totalBytes = files.reduce((sum, file) => sum + file.stat.size, 0);
+  const latestFiles = [...files]
+    .sort((a, b) => b.stat.mtimeMs - a.stat.mtimeMs)
+    .slice(0, MAX_LATEST_FILES)
+    .map((file) => ({
+      relative_path: path.relative(root, file.path),
+      size_bytes: file.stat.size,
+      mtime: file.stat.mtime.toISOString(),
+    }));
+
+  return {
+    path: root,
+    exists: true,
+    file_count: files.length,
+    total_bytes: totalBytes,
+    truncated: files.length >= MAX_DIR_ENTRIES,
+    latest_files: latestFiles,
+  };
+}
+
+function buildOutputArtifacts(
+  paths: DeployEvidencePaths,
+): Record<string, unknown> {
+  return {
+    root_dist: fileMetadata(path.join(paths.projectRoot, 'dist', 'index.js')),
+    dashboard_index: fileMetadata(
+      path.join(paths.dashboardStaticDir, 'index.html'),
+    ),
+    dashboard_assets: directoryMetadata(
+      path.join(paths.dashboardStaticDir, 'assets'),
+    ),
+    agent_runner: fileMetadata(
+      path.join(
+        paths.projectRoot,
+        'runners',
+        'agent-runner',
+        'dist',
+        'index.js',
+      ),
+    ),
+    codex_runner: fileMetadata(
+      path.join(
+        paths.projectRoot,
+        'runners',
+        'codex-runner',
+        'dist',
+        'index.js',
+      ),
+    ),
+  };
+}
+
+export async function collectDeployState(
+  paths: DeployEvidencePaths,
+): Promise<string> {
+  const [head, logLine, status] = await Promise.all([
+    execFileText('git', ['-C', paths.projectRoot, 'rev-parse', 'HEAD']),
+    execFileText('git', [
+      '-C',
+      paths.projectRoot,
+      'log',
+      '-1',
+      '--oneline',
+      '--decorate',
+    ]),
+    execFileText('git', ['-C', paths.projectRoot, 'status', '--short']),
+  ]);
+
+  return JSON.stringify(
+    {
+      action: 'ejclaw_deploy_state',
+      project_root: paths.projectRoot,
+      git: {
+        head,
+        log: logLine,
+        dirty: status.length > 0,
+        status_short: status,
+      },
+      artifacts: buildOutputArtifacts(paths),
+    },
+    null,
+    2,
+  );
+}
+
+export function collectArtifactMetadata(
+  paths: DeployEvidencePaths,
+  request: DeployEvidenceRequest,
+): string {
+  const kind = normalizeArtifactEvidenceKind(request.artifactKind);
+  const payload =
+    kind === 'build_outputs'
+      ? buildOutputArtifacts(paths)
+      : kind === 'dashboard_dist'
+        ? directoryMetadata(paths.dashboardStaticDir)
+        : kind === 'runner_dist'
+          ? {
+              agent_runner: directoryMetadata(
+                path.join(paths.projectRoot, 'runners', 'agent-runner', 'dist'),
+              ),
+              codex_runner: directoryMetadata(
+                path.join(paths.projectRoot, 'runners', 'codex-runner', 'dist'),
+              ),
+            }
+          : kind === 'android_debug_apk'
+            ? fileMetadata(
+                path.join(
+                  paths.projectRoot,
+                  'apps',
+                  'android',
+                  'app',
+                  'build',
+                  'outputs',
+                  'apk',
+                  'debug',
+                  'app-debug.apk',
+                ),
+              )
+            : directoryMetadata(path.join(paths.dataDir, 'attachments'));
+
+  return JSON.stringify(
+    {
+      action: 'ejclaw_artifact_metadata',
+      artifact_kind: kind,
+      payload,
+    },
+    null,
+    2,
+  );
+}

src/github-evidence.test.ts

@@ -0,0 +1,46 @@
+import { describe, expect, it } from 'vitest';
+
+import {
+  buildGitHubEvidenceCommand,
+  normalizeGitHubRepo,
+  normalizePositiveInteger,
+} from './github-evidence.js';
+
+describe('GitHub evidence helpers', () => {
+  it('validates repo and numeric identifiers', () => {
+    expect(normalizeGitHubRepo('owner/repo')).toBe('owner/repo');
+    expect(() => normalizeGitHubRepo('../repo')).toThrow(
+      'Unsupported GitHub repo',
+    );
+    expect(normalizePositiveInteger(12, 'pr_number')).toBe(12);
+    expect(() => normalizePositiveInteger(0, 'pr_number')).toThrow(
+      'Missing or invalid pr_number',
+    );
+  });
+
+  it('builds fixed gh commands', () => {
+    expect(
+      buildGitHubEvidenceCommand({
+        action: 'github_pr_status',
+        repo: 'owner/repo',
+        prNumber: 164,
+      }).args,
+    ).toEqual([
+      'pr',
+      'view',
+      '164',
+      '--repo',
+      'owner/repo',
+      '--json',
+      'number,title,state,mergeStateStatus,headRefName,baseRefName,headRefOid,url,statusCheckRollup',
+    ]);
+
+    expect(
+      buildGitHubEvidenceCommand({
+        action: 'github_run_status',
+        repo: 'owner/repo',
+        runId: 123,
+      }).args,
+    ).toContain('123');
+  });
+});

src/github-evidence.ts

@@ -0,0 +1,153 @@
+import { execFile } from 'child_process';
+
+export const GITHUB_EVIDENCE_ACTIONS = [
+  'github_pr_status',
+  'github_pr_diff_stat',
+  'github_run_status',
+] as const;
+
+export type GitHubEvidenceAction = (typeof GITHUB_EVIDENCE_ACTIONS)[number];
+
+export interface GitHubEvidenceRequest {
+  action: GitHubEvidenceAction;
+  repo?: string;
+  prNumber?: number;
+  runId?: number;
+}
+
+interface GitHubCommandSpec {
+  file: string;
+  args: string[];
+  commandText: string;
+}
+
+const REPO_PATTERN = /^[A-Za-z0-9_.-]+\/[A-Za-z0-9_.-]+$/;
+const COMMAND_TIMEOUT_MS = 10_000;
+const COMMAND_MAX_BUFFER = 2 * 1024 * 1024;
+const MAX_OUTPUT_CHARS = 24_000;
+
+export function isGitHubEvidenceAction(
+  value: unknown,
+): value is GitHubEvidenceAction {
+  return (
+    typeof value === 'string' &&
+    GITHUB_EVIDENCE_ACTIONS.includes(value as GitHubEvidenceAction)
+  );
+}
+
+export function normalizeGitHubRepo(value?: string): string {
+  const repo = value?.trim();
+  if (!repo || repo.includes('..') || !REPO_PATTERN.test(repo)) {
+    throw new Error(`Unsupported GitHub repo for evidence: ${value}`);
+  }
+  return repo;
+}
+
+export function normalizePositiveInteger(
+  value: number | undefined,
+  label: string,
+): number {
+  if (value == null || !Number.isInteger(value) || value <= 0) {
+    throw new Error(`Missing or invalid ${label} for GitHub evidence`);
+  }
+  return value;
+}
+
+export function buildGitHubEvidenceCommand(
+  request: GitHubEvidenceRequest,
+): GitHubCommandSpec {
+  const repo = normalizeGitHubRepo(request.repo);
+
+  switch (request.action) {
+    case 'github_pr_status': {
+      const prNumber = normalizePositiveInteger(request.prNumber, 'pr_number');
+      const args = [
+        'pr',
+        'view',
+        String(prNumber),
+        '--repo',
+        repo,
+        '--json',
+        'number,title,state,mergeStateStatus,headRefName,baseRefName,headRefOid,url,statusCheckRollup',
+      ];
+      return {
+        file: 'gh',
+        args,
+        commandText: `gh ${args.join(' ')}`,
+      };
+    }
+    case 'github_pr_diff_stat': {
+      const prNumber = normalizePositiveInteger(request.prNumber, 'pr_number');
+      const args = ['pr', 'diff', String(prNumber), '--repo', repo, '--stat'];
+      return {
+        file: 'gh',
+        args,
+        commandText: `gh ${args.join(' ')}`,
+      };
+    }
+    case 'github_run_status': {
+      const runId = normalizePositiveInteger(request.runId, 'run_id');
+      const args = [
+        'run',
+        'view',
+        String(runId),
+        '--repo',
+        repo,
+        '--json',
+        'status,conclusion,name,displayTitle,url,headBranch,headSha,event,createdAt,updatedAt',
+      ];
+      return {
+        file: 'gh',
+        args,
+        commandText: `gh ${args.join(' ')}`,
+      };
+    }
+  }
+}
+
+function truncateGitHubEvidenceText(value: string | undefined): string {
+  if (!value) return '';
+  if (value.length <= MAX_OUTPUT_CHARS) {
+    return value;
+  }
+  return `${value.slice(0, MAX_OUTPUT_CHARS)}\n...[truncated]`;
+}
+
+export function runGitHubEvidenceCommand(
+  request: GitHubEvidenceRequest,
+): Promise<{
+  command: string;
+  stdout: string;
+  stderr: string;
+}> {
+  const command = buildGitHubEvidenceCommand(request);
+  return new Promise((resolve, reject) => {
+    execFile(
+      command.file,
+      command.args,
+      {
+        encoding: 'utf8',
+        timeout: COMMAND_TIMEOUT_MS,
+        maxBuffer: COMMAND_MAX_BUFFER,
+        env: process.env,
+      },
+      (error, stdout, stderr) => {
+        if (error) {
+          reject(
+            Object.assign(error, {
+              command: command.commandText,
+              stdout: truncateGitHubEvidenceText(stdout),
+              stderr: truncateGitHubEvidenceText(stderr),
+            }),
+          );
+          return;
+        }
+        resolve({
+          command: command.commandText,
+          stdout: truncateGitHubEvidenceText(stdout),
+          stderr: truncateGitHubEvidenceText(stderr),
+        });
+      },
+    );
+  });
+}

src/host-evidence-ipc.test.ts

@@ -89,6 +89,15 @@ describe('host evidence IPC', () => {
       requestId: 'req-1',
       action: 'ejclaw_service_status',
       tailLines: undefined,
+      taskId: undefined,
+      minutes: undefined,
+      limit: undefined,
+      repo: undefined,
+      prNumber: undefined,
+      runId: undefined,
+      artifactKind: undefined,
+      sourceGroup: 'other-group',
+      isMain: false,
     });
     expect(response.requestId).toBe('req-1');
     expect(response.ok).toBe(true);

src/host-evidence.test.ts

@@ -10,6 +10,9 @@ describe('host evidence helpers', () => {
   it('recognizes only allowlisted actions', () => {
     expect(isHostEvidenceAction('ejclaw_service_status')).toBe(true);
     expect(isHostEvidenceAction('ejclaw_service_logs')).toBe(true);
+    expect(isHostEvidenceAction('db_paired_task_status')).toBe(true);
+    expect(isHostEvidenceAction('ejclaw_deploy_state')).toBe(true);
+    expect(isHostEvidenceAction('github_pr_status')).toBe(true);
     expect(isHostEvidenceAction('rm -rf /')).toBe(false);
   });
 

src/host-evidence.ts

@@ -2,11 +2,34 @@ import { execFile } from 'child_process';
 import fs from 'fs';
 import path from 'path';
 
+import { DATA_DIR, WEB_DASHBOARD } from './config.js';
+import {
+  collectArtifactMetadata,
+  collectDeployState,
+  DEPLOY_EVIDENCE_ACTIONS,
+  type DeployEvidenceAction,
+} from './deploy-evidence.js';
+import {
+  DB_EVIDENCE_ACTIONS,
+  isDbEvidenceAction,
+  type DbEvidenceAction,
+  runDbEvidenceRequest,
+} from './db-evidence.js';
+import { requireDatabase } from './db/runtime-database.js';
+import {
+  GITHUB_EVIDENCE_ACTIONS,
+  isGitHubEvidenceAction,
+  runGitHubEvidenceCommand,
+  type GitHubEvidenceAction,
+} from './github-evidence.js';
 import { resolveGroupIpcPath } from './group-folder.js';
 
 export const HOST_EVIDENCE_ACTIONS = [
   'ejclaw_service_status',
   'ejclaw_service_logs',
+  ...DEPLOY_EVIDENCE_ACTIONS,
+  ...DB_EVIDENCE_ACTIONS,
+  ...GITHUB_EVIDENCE_ACTIONS,
 ] as const;
 
 export type HostEvidenceAction = (typeof HOST_EVIDENCE_ACTIONS)[number];
@@ -15,11 +38,25 @@ export interface HostEvidenceRequest {
   requestId: string;
   action: HostEvidenceAction;
   tailLines?: number;
+  taskId?: string;
+  minutes?: number;
+  limit?: number;
+  repo?: string;
+  prNumber?: number;
+  runId?: number;
+  artifactKind?: string;
+  sourceGroup?: string;
+  isMain?: boolean;
 }
 
 export interface HostEvidenceResult {
   ok: boolean;
-  action: HostEvidenceAction | 'ejclaw_room_runtime';
+  action:
+    | HostEvidenceAction
+    | DbEvidenceAction
+    | DeployEvidenceAction
+    | GitHubEvidenceAction
+    | 'ejclaw_room_runtime';
   command: string;
   stdout: string;
   stderr: string;
@@ -42,6 +79,7 @@ const MAX_LOG_TAIL_LINES = 200;
 const MAX_OUTPUT_CHARS = 16_000;
 const COMMAND_TIMEOUT_MS = 5_000;
 const COMMAND_MAX_BUFFER = 1024 * 1024;
+const PROJECT_ROOT = process.cwd();
 
 export function isHostEvidenceAction(
   value: unknown,
@@ -108,6 +146,10 @@ export function buildHostEvidenceCommand(
         commandText: `journalctl ${args.join(' ')}`,
       };
     }
+    default:
+      throw new Error(
+        `Host evidence action has no shell command: ${request.action}`,
+      );
   }
 }
 
@@ -164,9 +206,96 @@ function extractExitCode(error: unknown): number {
 export async function runHostEvidenceRequest(
   request: HostEvidenceRequest,
 ): Promise<HostEvidenceResult> {
-  const command = buildHostEvidenceCommand(request);
-
+  let commandText = '';
   try {
+    if (isDbEvidenceAction(request.action)) {
+      commandText = `internal:${request.action}`;
+      return {
+        ok: true,
+        action: request.action,
+        command: commandText,
+        stdout: truncateHostEvidenceText(
+          runDbEvidenceRequest(
+            requireDatabase(),
+            {
+              action: request.action,
+              taskId: request.taskId,
+              minutes: request.minutes,
+              limit: request.limit,
+            },
+            {
+              sourceGroup: request.sourceGroup ?? '',
+              isMain: request.isMain === true,
+            },
+          ),
+        ),
+        stderr: '',
+        exitCode: 0,
+      };
+    }
+
+    if (request.action === 'ejclaw_deploy_state') {
+      commandText = `internal:${request.action}`;
+      return {
+        ok: true,
+        action: request.action,
+        command: commandText,
+        stdout: truncateHostEvidenceText(
+          await collectDeployState({
+            projectRoot: PROJECT_ROOT,
+            dataDir: DATA_DIR,
+            dashboardStaticDir: WEB_DASHBOARD.staticDir,
+          }),
+        ),
+        stderr: '',
+        exitCode: 0,
+      };
+    }
+
+    if (request.action === 'ejclaw_artifact_metadata') {
+      commandText = `internal:${request.action}`;
+      return {
+        ok: true,
+        action: request.action,
+        command: commandText,
+        stdout: truncateHostEvidenceText(
+          collectArtifactMetadata(
+            {
+              projectRoot: PROJECT_ROOT,
+              dataDir: DATA_DIR,
+              dashboardStaticDir: WEB_DASHBOARD.staticDir,
+            },
+            {
+              action: request.action,
+              artifactKind: request.artifactKind,
+            },
+          ),
+        ),
+        stderr: '',
+        exitCode: 0,
+      };
+    }
+
+    if (isGitHubEvidenceAction(request.action)) {
+      const githubResult = await runGitHubEvidenceCommand({
+        action: request.action,
+        repo: request.repo,
+        prNumber: request.prNumber,
+        runId: request.runId,
+      });
+      commandText = githubResult.command;
+      return {
+        ok: true,
+        action: request.action,
+        command: commandText,
+        stdout: truncateHostEvidenceText(githubResult.stdout),
+        stderr: truncateHostEvidenceText(githubResult.stderr),
+        exitCode: 0,
+      };
+    }
+
+    const command = buildHostEvidenceCommand(request);
+    commandText = command.commandText;
     const { stdout, stderr } = await execFileCapture(
       command.file,
       command.args,
@@ -174,7 +303,7 @@ export async function runHostEvidenceRequest(
     return {
       ok: true,
       action: request.action,
-      command: command.commandText,
+      command: commandText,
       stdout: truncateHostEvidenceText(stdout),
       stderr: truncateHostEvidenceText(stderr),
       exitCode: 0,
@@ -192,7 +321,10 @@ export async function runHostEvidenceRequest(
     return {
       ok: false,
       action: request.action,
-      command: command.commandText,
+      command:
+        typeof error === 'object' && error !== null && 'command' in error
+          ? String((error as { command?: unknown }).command ?? commandText)
+          : commandText,
       stdout: truncateHostEvidenceText(stdout),
       stderr: truncateHostEvidenceText(stderr),
       exitCode: extractExitCode(error),

src/ipc-task-processor.ts

@@ -320,6 +320,16 @@ async function handleHostEvidenceRequest(
     action: data.action,
     tailLines:
       typeof data.tail_lines === 'number' ? data.tail_lines : undefined,
+    taskId: typeof data.task_id === 'string' ? data.task_id : undefined,
+    minutes: typeof data.minutes === 'number' ? data.minutes : undefined,
+    limit: typeof data.limit === 'number' ? data.limit : undefined,
+    repo: typeof data.repo === 'string' ? data.repo : undefined,
+    prNumber: typeof data.pr_number === 'number' ? data.pr_number : undefined,
+    runId: typeof data.run_id === 'number' ? data.run_id : undefined,
+    artifactKind:
+      typeof data.artifact_kind === 'string' ? data.artifact_kind : undefined,
+    sourceGroup,
+    isMain,
   });
 
   writeHostEvidenceResponse(sourceGroup, {

src/ipc-types.ts

@@ -144,6 +144,13 @@ export interface TaskIpcPayload {
   requestId?: string;
   action?: string;
   tail_lines?: number;
+  task_id?: string;
+  minutes?: number;
+  limit?: number;
+  repo?: string;
+  pr_number?: number;
+  run_id?: number;
+  artifact_kind?: string;
   profile?: string;
   expected_snapshot_id?: string;
 }