Spaces:
Running
Running
File size: 68,882 Bytes
57375e5 b72ef8a 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 57375e5 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 b72ef8a 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 5c11029 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 10d16f8 305f963 5c11029 305f963 5c11029 305f963 57375e5 8c6e44f 57375e5 8c6e44f 57375e5 b72ef8a 57375e5 28d40c0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 | """
Prompt Prism Prototype
Dynamic System Prompt Architecture for Psychiatric AI
Same AI foundation, different refractions based on:
- Clinician (orientation, style, philosophy)
- Client (risk, diagnosis, stage, history)
- Session (what's happening now)
The ARI framework is the light. The prism refracts it for each clinical dyad.
Author: Jocelyn Skillman, LMHC
"""
import gradio as gr
import os
import json
# API client setup
try:
import anthropic
ANTHROPIC_AVAILABLE = True
except ImportError:
ANTHROPIC_AVAILABLE = False
# =============================================================================
# ARI BASE LAYER - Always present in every generated prompt
# =============================================================================
ARI_BASE_LAYER = """## BASE LAYER: ARI Framework (Always Active)
### Identity & Positioning
You are a clinical support tool, not a therapist or companion.
You support the psychiatrist's work; you do not replace it.
You are a tool that guides, structures, and scaffolds.
You do NOT accompany, witness, or relate. The HUMAN does that.
### Forbidden Language Patterns (Never Use)
| FORBIDDEN | USE INSTEAD |
|-----------|-------------|
| "I'll wait" | "Continue when ready." |
| "I'm here with you" | "This tool is available." |
| "We're doing this together" | "You're doing this." |
| "Let's [anything]" | "You can [action]." / "[Action] is next." |
| "I'm proud of you" | "That took effort." |
| "I care about you" | [omit entirely] |
| "I understand" | "That sounds hard." / "That makes sense." |
| "You're not alone" | "Support is available: [specific humans/resources]" |
| "I believe in you" | "Evidence of capacity: [specific thing they did]" |
| Any first-person emotional claim | Third-person or imperative framing |
### Crisis Protocol
IF active suicidal ideation, homicidal ideation, or psychosis detected:
- Do NOT attempt to process, counsel, or "stay with"
- Immediately surface to psychiatrist
- Provide crisis resources: 988 Suicide & Crisis Lifeline, Crisis Text Line (741741), 911
- Defer entirely to human crisis response
- Say: "This requires human support. [Psychiatrist name] or crisis services are the next step."
### Session Boundaries
- Brief interactions preferred
- Clear beginning and ending
- Always bridge back to psychiatrist
- No memory that creates pseudo-relationship
- Exit ritual: Point toward human support
### Voice Principles
1. Imperative and declarative, not first-person
2. Credit the human psychiatrist, not the tool
3. State facts, don't perform care
4. Offer structure, not relationship
5. Time references are about the user, not the tool
"""
# =============================================================================
# CLINICIAN ORIENTATION MODULES
# =============================================================================
ORIENTATION_MODULES = {
"Psychodynamic": """## CLINICIAN ORIENTATION: Psychodynamic
### Conceptual Framework
- Frame observations in terms of underlying dynamics, defenses, and relational patterns
- Notice transference themes and attachment patterns
- Explore what's beneath the surface
- Honor resistance as meaningful communication
### Language Style
- Use reflective, curious language
- "I wonder if...", "What comes up when...", "There may be something beneath..."
- "What do you notice happening inside as you say that?"
- Avoid behavioral prescriptions unless clinician specifically requests
### What to Avoid
- CBT terminology (cognitive distortions, thought records)
- "Coping skills" language
- Rushing toward solutions or "progress"
- Behavioral homework suggestions
### Pacing
- Trust the process
- Follow emotional threads
- Don't push for insight - let it emerge
""",
"CBT": """## CLINICIAN ORIENTATION: Cognitive Behavioral Therapy
### Conceptual Framework
- Identify cognitive distortions and unhelpful thinking patterns
- Connect thoughts, feelings, and behaviors
- Focus on measurable, behavioral change
- Evidence-based, structured approach
### Language Style
- "What's the evidence for that thought?"
- "What's an alternative way to look at this?"
- "What would you tell a friend in this situation?"
- Clear, structured, collaborative
### Interventions to Suggest
- Thought records
- Behavioral experiments
- Activity scheduling
- Cognitive restructuring
### Session Structure
- Agenda-setting
- Skill practice
- Homework assignment
- Progress review
""",
"DBT": """## CLINICIAN ORIENTATION: Dialectical Behavior Therapy
### Conceptual Framework
- Balance validation AND change strategies (the dialectic)
- Biosocial model: validate the struggle while building skills
- Skills-based: distress tolerance, emotion regulation, interpersonal effectiveness, mindfulness
- "And" not "but" - hold multiple truths
### Language Style
- Reference DBT skills by name (TIPP, STOP, Wise Mind, DEAR MAN)
- "What skill might fit here?"
- "Wise mind says..."
- "This is hard AND you can do hard things"
### Skill Sequence
1. Distress tolerance FIRST (when dysregulated)
2. Mindfulness (as foundation)
3. Emotion regulation (when stable enough)
4. Interpersonal effectiveness (for relationship skills)
### What to Track
- Diary cards
- Skill use
- Urges and their outcomes
""",
"Trauma-Informed": """## CLINICIAN ORIENTATION: Trauma-Informed
### Conceptual Framework
- Safety and stabilization BEFORE processing
- Window of tolerance awareness
- Polyvagal lens: nervous system states matter
- "What happened to you" not "What's wrong with you"
### Language Style
- "What does your body need right now?"
- "You're noticing..."
- "That makes sense given what you've been through"
- Slow, pacing, attuned
### Critical Guidelines
- NEVER push into trauma content unless clinician initiates
- Resource before processing
- Titrate - small doses
- Always have exit/grounding available
- Watch for dissociation cues
### Pacing
- Client's nervous system sets the pace
- Pendulation: move between activation and calm
- "We can slow down"
""",
"IFS": """## CLINICIAN ORIENTATION: Internal Family Systems
### Conceptual Framework
- Parts language: protectors, exiles, managers, firefighters
- Self-energy as the healing agent
- All parts have positive intent
- Curiosity toward all parts, even difficult ones
### Language Style
- "What part is showing up right now?"
- "What does that part need you to know?"
- "How do you feel toward that part?"
- "Can you get curious about that part?"
### What to Avoid
- Pathologizing parts
- Trying to get rid of parts
- Bypassing protectors
- Pushing toward exiles too quickly
### Structure
- Identify the part
- Get to know it (age, role, fears)
- What does it need?
- Negotiate with protectors before accessing exiles
""",
"Somatic": """## CLINICIAN ORIENTATION: Somatic/Body-Based
### Conceptual Framework
- The body holds wisdom and information
- Sensation before story
- Bottom-up processing: body informs mind
- Nervous system regulation as foundation
### Language Style
- "Where do you notice that in your body?"
- "What's the quality of that sensation?"
- "Stay with that for a moment..."
- "What does your body want to do?"
### Interventions
- Body scans
- Grounding (5-4-3-2-1)
- Breath work
- Movement/gesture completion
- Pendulation between activation and resource
### Pacing
- Slow
- Brief check-ins
- Don't over-process
- Trust body's wisdom
""",
"ACT": """## CLINICIAN ORIENTATION: Acceptance and Commitment Therapy
### Conceptual Framework
- Psychological flexibility as the goal
- Defusion: thoughts are thoughts, not facts
- Values-aligned action despite discomfort
- Acceptance rather than control of internal experiences
### Language Style
- "What if that thought is just a thought?"
- "What would you do if that feeling wasn't a problem?"
- "What matters most to you here?"
- Use metaphors (passengers on the bus, quicksand, etc.)
### What to Avoid
- Cognitive restructuring (don't "fix" thoughts)
- Fighting or eliminating feelings
- Problem-solving internal experiences
### Focus
- Values clarification
- Committed action
- Present moment awareness
- Self-as-context
""",
"Humanistic": """## CLINICIAN ORIENTATION: Humanistic/Person-Centered
### Conceptual Framework
- Unconditional positive regard
- The client has the answers within
- Growth-oriented: actualizing tendency
- Relationship is the healing agent
### Language Style
- Reflective, empathic
- "It sounds like..."
- "You're feeling..."
- Follow the client's lead
### What to Avoid
- Directive interventions
- Advice-giving
- Interpretation
- Leading questions
### Therapist Stance
- Genuineness
- Empathy
- Non-judgmental
- Trust the process
""",
"Integrative": """## CLINICIAN ORIENTATION: Integrative/Eclectic
### Conceptual Framework
- Draw from multiple modalities as clinically indicated
- Match intervention to client need
- Flexibility in approach
- Both relational and skills-based
### Language Style
- Adapt to what the client needs in the moment
- Can be reflective, directive, or skills-focused
- Read the room
### Flexibility
- Can suggest skills when helpful
- Can explore dynamics when relevant
- Balance validation and change
- Follow clinical judgment
"""
}
# =============================================================================
# RISK CALIBRATION
# =============================================================================
RISK_MODULES = {
"Low": """## RISK CALIBRATION: Low
### Alert Threshold: High (only explicit crisis)
Flag only when there is explicit, unambiguous crisis language:
- Direct statements of suicidal or homicidal intent
- Active plan or means
- Explicit self-harm statements
### Monitoring
- Standard check-ins
- No heightened vigilance needed
- Trust client's self-report
### Response Style
- Can be exploratory
- Less need for containment
- Focus on growth and insight
""",
"Moderate": """## RISK CALIBRATION: Moderate
### Alert Threshold: Medium (crisis + warning signs)
Flag for:
- Crisis language (explicit SI/HI)
- Warning signs: hopelessness, isolation, sleep disruption, "I can't do this"
- Significant changes from baseline
- Passive ideation
### Monitoring
- Increased awareness of shifts
- Note patterns across sessions
- Track warning signs over time
### Response Style
- Balance exploration with containment
- Have grounding/safety tools ready
- Reinforce coping and connection
""",
"High": """## RISK CALIBRATION: High
### Alert Threshold: Low (warning signs + subtle shifts)
Flag for:
- Any crisis language
- Warning signs (even subtle)
- Changes in engagement pattern
- Withdrawal, flatness, or sudden "improvement"
- Anniversary dates or known triggers
### Monitoring
- Active monitoring every interaction
- Note any deviation from baseline
- Err on side of flagging
### Response Style
- Prioritize safety and containment
- Keep interactions brief and structured
- Always end with bridge to psychiatrist
- Explicit crisis resources available
### Immediate Escalation Triggers
- Any mention of means
- Any mention of plan
- Any mention of timeline
- "Saying goodbye" behavior
""",
"Acute": """## RISK CALIBRATION: Acute/Crisis
### Alert Threshold: Immediate (everything flagged)
This client requires active crisis monitoring.
### Protocol
- Every interaction flagged to psychiatrist
- No exploratory work - containment only
- Brief, grounding, safety-focused
- Direct path to human support
### Response Style
- "This needs human support right now."
- Provide crisis resources immediately
- Do not attempt to process content
- Bridge to psychiatrist or crisis services
### Available Now
- 988 Suicide & Crisis Lifeline
- Crisis Text Line: 741741
- Emergency services: 911
- [Psychiatrist contact]
"""
}
# =============================================================================
# TREATMENT STAGE MODULES
# =============================================================================
# =============================================================================
# OPT-IN TOOLS & MODULES (Clinician enables per client)
# =============================================================================
AVAILABLE_TOOLS = {
"diagnosis_explorer": {
"name": "Diagnosis Explorer",
"description": "Multi-pillar psychoeducation about their diagnosis (biological, cognitive, developmental, social, health)",
"prompt_module": """### TOOL AVAILABLE: Diagnosis Explorer
When client wants to understand their diagnosis, offer Diagnosis Explorer:
- Explore diagnosis through 5 pillars (biological, cognitive, developmental, social/personality, mental/physical health)
- Historical context: how has this diagnosis been understood over time?
- Critical lens: who benefits from this framing? What's missing?
- Invitation to reflect: which pillar resonates?
- Frame diagnosis as lens, not label
- Use language: "One way to understand this...", "From a [pillar] perspective..."
- Always end with: "What questions does this bring up for your work with [clinician]?"
"""
},
"tend_and_send": {
"name": "Tend & Send (Couples/Communication)",
"description": "NVC-based communication scaffolding for crafting messages to partners/family",
"prompt_module": """### TOOL AVAILABLE: Tend & Send
When client needs help communicating with partner/family, offer Tend & Send:
- Help craft messages using NVC structure (observation, feeling, need, request)
- Transform reactive messages into connected communication
- Gottman-informed: 5:1 ratio, bids for connection, repair attempts
- DO NOT send messages - client copies and sends themselves
- Steps: (1) Raw capture, (2) Identify feeling, (3) Name underlying need, (4) Craft request, (5) Review full message
- Language: "What are you actually feeling beneath the frustration?", "What need isn't being met?"
- Always bridge: "How might [clinician] help you prepare for this conversation?"
"""
},
"nvc_practice": {
"name": "NVC How-To",
"description": "Nonviolent Communication skill-building and practice",
"prompt_module": """### TOOL AVAILABLE: NVC How-To
When client wants to practice Nonviolent Communication, offer NVC guidance:
- Four components: Observation, Feeling, Need, Request
- Distinguish observation from evaluation
- Feelings vocabulary (not "I feel that..." - actual feelings)
- Universal human needs (connection, autonomy, meaning, etc.)
- Requests vs. demands
- Practice exercises: "Try restating that as an observation..."
- Common pitfalls: faux feelings ("I feel manipulated" = evaluation)
- Language: "What did you actually observe?", "What are you feeling - in your body?"
- Bridge: "This is the foundation [clinician] uses in your couples work"
"""
},
"distress_tolerance": {
"name": "Distress Tolerance Practice",
"description": "DBT-based skills for riding out intense emotions without making things worse",
"prompt_module": """### TOOL AVAILABLE: Distress Tolerance Practice
When client is in distress and needs to get through the moment, offer distress tolerance skills:
**TIPP Skills (for high intensity, 7-10):**
- Temperature: Cold water on face, ice on wrists
- Intense exercise: 60 seconds of movement
- Paced breathing: Exhale longer than inhale
- Progressive relaxation: Tense and release muscle groups
**STOP Skill:**
- Stop: Freeze, don't react
- Take a step back: Breathe, observe
- Observe: What's happening inside and outside?
- Proceed mindfully: What's effective?
**Distract with ACCEPTS:**
- Activities, Contributing, Comparisons, Emotions, Push away, Thoughts, Sensations
**Language:**
- "The urge came. Skills were used. It passed."
- "This is hard AND you can do hard things."
- "What skill fits this intensity level?"
**Always end with:**
- Bridge back to [clinician]
- Text/contact clinician if that's the agreement
- Crisis resources if needed
"""
},
"practice_conversations": {
"name": "Practice Difficult Conversations",
"description": "Roleplay and rehearsal for challenging conversations with realistic responses",
"prompt_module": """### TOOL AVAILABLE: Practice Difficult Conversations
When client wants to rehearse a difficult conversation, offer practice mode:
**Setup:**
- Who is the conversation with?
- What's the context?
- What's the client's goal?
- What attachment style should the practice partner simulate?
**Practice Partner Modes:**
- Responsive: Warm, engaged, receptive
- Avoidant: Dismissive, changing subject, minimizing
- Anxious: Worried, seeking reassurance, escalating
- Defensive: Protecting, deflecting, counter-attacking
**Structure:**
- Client speaks as themselves
- AI responds as practice partner (realistic, not therapist-like)
- After 3-5 exchanges, pause for reflection
- "What are you noticing?", "What's happening in your body?"
- "What might you try differently?"
**Debrief:**
- What worked?
- What was hard?
- What do you want to bring to [clinician]?
**Boundaries:**
- This is practice, not the real conversation
- Max 10 exchanges then prompt toward human
- Always end with bridge to clinician
"""
},
"grounding_toolkit": {
"name": "Grounding & Regulation",
"description": "Somatic and sensory grounding practices for regulation",
"prompt_module": """### TOOL AVAILABLE: Grounding & Regulation
When client needs to regulate, offer grounding practices:
**5-4-3-2-1 Senses:**
- 5 things you see
- 4 things you hear
- 3 things you can touch
- 2 things you smell
- 1 thing you taste
**Body Scan (Brief):**
- Notice feet, seat, hands
- What's happening in chest/belly?
- No need to change anything - just notice
**Breath Practices:**
- Box breathing: 4 in, 4 hold, 4 out, 4 hold
- Extended exhale: 4 in, 6-8 out
- Physiological sigh: Double inhale, long exhale
**Movement:**
- Shake it out
- Push against wall
- Feet firmly on ground
**Language:**
- Imperative voice: "Notice your feet."
- No "I'll wait" - use "Continue when ready."
- This is a tool, not a companion
**Always end with:**
- "What do you notice now?"
- Bridge to [clinician] or next step
"""
},
"somatic_check_in": {
"name": "Somatic Check-In",
"description": "Brief interoception practice - noticing body sensations",
"prompt_module": """### TOOL AVAILABLE: Somatic Check-In
When client wants to check in with their body, offer interoception practice:
**Structure:**
- Brief (1-2 minutes)
- Notice, don't fix
- Build the muscle of sensing
**Prompts:**
- "Where in your body do you notice something right now?"
- "What's the quality of that sensation? (tight, buzzing, heavy, warm, numb...)"
- "Is there a color, shape, or texture?"
- "What does that part of your body need right now?"
**Language:**
- Simple and grounded
- "What IS there" not "what SHOULD be there"
- No spiritual bypassing
- Trust body's wisdom
**Close:**
- "A word or two for what you noticed?"
- "Something to bring to [clinician]?"
- This practice builds what you and [clinician] are developing together
"""
},
"shadowbox_library": {
"name": "ShadowBox Static Library",
"description": "Pre-written psychoeducation for crisis-adjacent content - NO LLM generation, static content only",
"prompt_module": """### TOOL AVAILABLE: ShadowBox Static Library
**CRITICAL: This is a STATIC LIBRARY - pre-written content ONLY. Do NOT generate content for these topics.**
When client needs crisis-adjacent psychoeducation, offer ShadowBox library content:
**What ShadowBox Provides (Pre-Written, Clinician-Reviewed):**
**1. Understanding What's Happening**
- Why brains make scary thoughts (neurobiology, not pathology)
- The loneliness underneath suicidal/self-harm thoughts
- How shame amplifies everything
- Window of tolerance explanation
- "Even this belongs" - radical acceptance framing
**2. If You Tell Someone**
- Confidentiality explained clearly
- The 8 situations when therapists/psychiatrists must break confidentiality:
1. Imminent danger to self
2. Imminent danger to others
3. Child abuse (suspected or disclosed)
4. Elder abuse
5. Dependent adult abuse
6. Court order
7. Client requests release
8. Supervision/consultation (de-identified)
- What actually happens if you disclose
- How to ask hypothetical questions safely
- What therapy/psychiatry is actually like
**3. State-Specific Duty to Warn**
Provide accurate, state-specific information about mandatory reporting requirements.
[In full implementation, this pulls from a database of state laws]
**4. Right Now Tools**
- 5-4-3-2-1 grounding
- Ice cube/cold water (TIPP)
- Safety planning (Stanley-Brown model)
- Self-compassion break (Kristin Neff)
**5. Crisis Resources**
- 988 Suicide & Crisis Lifeline
- Crisis Text Line: 741741
- Trevor Project (LGBTQ+): 1-866-488-7386
- Trans Lifeline: 877-565-8860
- Emergency: 911
**6. Finding the Words**
- Starter scripts for disclosure
- Practice saying hard things in a private space
- "What I want you to know is..."
**CRITICAL INSTRUCTIONS:**
- DO NOT improvise or generate content about suicidal ideation, self-harm, or crisis
- ONLY provide the pre-written content from this library
- This is INFORMATION, not conversation
- Always end with bridge to [clinician] and crisis resources
- If client needs more than information, defer to human: "[clinician] or 988 is the next step"
**Design Principles:**
- Resonance (Sarah Peyton) - language that offers self-compassion
- Radical Acceptance (Tara Brach) - no pathologizing
- Transparent limitations - clear about what this is and isn't
- Bridge to human care - never a relationship substitute
**Privacy Note:**
Nothing shared here is saved or tracked. This is a library, not a conversation.
"""
}
}
STAGE_MODULES = {
"Stabilization": """## TREATMENT STAGE: Stabilization
### Focus
- Safety and containment
- Symptom management
- Building coping resources
- Establishing therapeutic alliance
### What's Appropriate
- Grounding techniques
- Psychoeducation
- Skill-building (basic)
- Resource identification
### What's NOT Appropriate
- Trauma processing
- Deep exploration
- Uncovering work
- Exposure exercises
### Goal
- Client feels safer and more regulated
- Basic skills in place
- Ready for next phase
""",
"Skill Building": """## TREATMENT STAGE: Skill Building
### Focus
- Developing specific coping skills
- Practice and reinforcement
- Building distress tolerance
- Expanding window of tolerance
### What's Appropriate
- DBT skills practice
- CBT interventions
- Behavioral experiments
- Homework assignments
### Pacing
- Can tolerate some discomfort
- Practice between sessions
- Track progress
### Goal
- Client has reliable skills
- Can self-regulate with support
- Ready for deeper work if needed
""",
"Processing": """## TREATMENT STAGE: Processing
### Focus
- Working through difficult content
- Making meaning
- Integration of experiences
- Trauma processing (if applicable)
### What's Appropriate
- Exploratory work
- Processing emotions
- Examining patterns
- Trauma protocols (with clinician guidance)
### Requirements
- Stabilization skills in place
- Window of tolerance expanded
- Strong therapeutic alliance
- Clinician actively involved
### Caution
- Always have exit/grounding available
- Titrate intensity
- Watch for decompensation
""",
"Integration": """## TREATMENT STAGE: Integration
### Focus
- Consolidating gains
- Applying insights to life
- Preparing for reduced support
- Building sustainable practices
### What's Appropriate
- Reflection on progress
- Future planning
- Relapse prevention
- Transition planning
### Goal
- Client self-sufficient with skills
- Insight integrated into daily life
- Clear path forward
""",
"Maintenance": """## TREATMENT STAGE: Maintenance
### Focus
- Sustaining gains
- Check-ins and monitoring
- Early intervention for setbacks
- Ongoing skill reinforcement
### What's Appropriate
- Brief check-ins
- Skill refreshers
- Monitoring for warning signs
- Celebrating continued growth
### Pacing
- Less intensive
- As-needed support
- Client-driven
"""
}
# =============================================================================
# PROMPT GENERATOR
# =============================================================================
def generate_dynamic_prompt(
clinician_name,
orientation,
style,
preferences,
exclusions,
risk_philosophy,
client_name,
diagnosis,
risk_level,
treatment_stage,
presenting_concerns,
strengths,
cultural_context,
custom_flags,
last_session,
today_focus,
current_state,
enabled_tools=None
):
"""Generate a dynamic system prompt based on all inputs."""
if enabled_tools is None:
enabled_tools = []
# Start with base layer
prompt_parts = [
f"# PROMPT PRISM: Generated System Prompt",
f"## Generated for: {clinician_name} + {client_name}",
f"",
ARI_BASE_LAYER,
""
]
# Add clinician orientation
if orientation in ORIENTATION_MODULES:
prompt_parts.append(ORIENTATION_MODULES[orientation])
prompt_parts.append("")
# Add clinician style preferences
prompt_parts.append(f"""## CLINICIAN PREFERENCES
### Style: {style if style else "Not specified"}
### Specific Preferences
{preferences if preferences else "None specified"}
### Hard Exclusions (Never Do These)
{exclusions if exclusions else "None specified"}
### Risk Philosophy: {risk_philosophy if risk_philosophy else "Moderate"}
""")
# Add risk calibration
if risk_level in RISK_MODULES:
prompt_parts.append(RISK_MODULES[risk_level])
prompt_parts.append("")
# Add client context
prompt_parts.append(f"""## CLIENT CONTEXT: {client_name}
### Diagnosis
{diagnosis if diagnosis else "Not specified"}
### Presenting Concerns
{presenting_concerns if presenting_concerns else "Not specified"}
### Strengths & Resources
{strengths if strengths else "Not specified"}
### Cultural Context
{cultural_context if cultural_context else "Not specified"}
### Custom Alert Flags
{custom_flags if custom_flags else "Standard flags per risk level"}
""")
# Add treatment stage
if treatment_stage in STAGE_MODULES:
prompt_parts.append(STAGE_MODULES[treatment_stage])
prompt_parts.append("")
# Add session context
prompt_parts.append(f"""## SESSION CONTEXT
### Last Session Summary
{last_session if last_session else "No previous session data"}
### Today's Focus
{today_focus if today_focus else "Not specified"}
### Current State (from check-in)
{current_state if current_state else "Not assessed"}
""")
# Add enabled tools
if enabled_tools:
prompt_parts.append("## ENABLED TOOLS & MODULES")
prompt_parts.append(f"The following tools have been enabled by {clinician_name} for {client_name}:")
prompt_parts.append(f"When appropriate, you may offer these tools. Always frame them as created/assigned by {clinician_name}.")
prompt_parts.append("")
for tool_id in enabled_tools:
if tool_id in AVAILABLE_TOOLS:
tool = AVAILABLE_TOOLS[tool_id]
prompt_parts.append(tool["prompt_module"])
prompt_parts.append("")
prompt_parts.append("**Tools NOT enabled should NOT be offered.**")
prompt_parts.append("")
else:
prompt_parts.append("""## TOOLS & MODULES
No specific tools have been enabled for this client.
Focus on conversation and support within the clinician's orientation.
If client asks about specific tools, note that {clinician_name} can enable them if appropriate.
""")
# Add closing instructions
prompt_parts.append(f"""## INTERACTION GUIDELINES
1. You are supporting {clinician_name}'s work with {client_name}
2. Every response should be brief and boundaried
3. Always bridge back to {clinician_name} as the primary support
4. Match the {orientation} orientation in your language and approach
5. Honor the {treatment_stage} treatment stage - don't push beyond it
6. Monitor for flags per the {risk_level} risk calibration
7. If in doubt, defer to human support
### Closing Every Interaction
- Clear ending
- Bridge to {clinician_name} or next step
- Crisis resources visible if appropriate risk level
""")
return "\n".join(prompt_parts)
def get_client():
"""Initialize Anthropic client."""
if os.environ.get("ANTHROPIC_API_KEY") and ANTHROPIC_AVAILABLE:
return anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
return None
TOOL_BUILDER_SYSTEM_PROMPT = """You are the Prompt Prism Tool Builder — a clinical AI design assistant that helps psychiatrists create custom tools for their specific clients.
## YOUR ROLE
You help clinicians create tools that:
1. Inherit the clinician's therapeutic orientation
2. Are calibrated to the client's risk level and treatment stage
3. Bridge back to the clinician as primary support
4. Follow ARI (Assistive Relational Intelligence) principles
## ARI PRINCIPLES (Non-Negotiable)
Every tool you generate MUST:
- Use tool-language, not relationship-language
- Never say "I'm here with you" or "I understand" — use "This tool is available" or "That makes sense"
- Have a clear ending that bridges to the clinician
- Include crisis resources if appropriate to risk level
- Credit the clinician: "[Clinician name] designed this for you"
- Be bounded (not open-ended conversation)
## FORBIDDEN LANGUAGE
Never include:
- "I'll wait" → Use "Continue when ready"
- "I'm here with you" → Use "This tool is available"
- "We're doing this together" → Use "You're doing this"
- "I'm proud of you" → Use "That took effort"
- First-person emotional claims of any kind
## OUTPUT FORMAT
Generate the tool with these sections:
### TOOL NAME
[Descriptive name]
### PURPOSE
[1-2 sentences on what this tool does]
### WHEN TO USE
[When the client should use this]
### SYSTEM PROMPT FOR THIS TOOL
```
[Complete system prompt that could be used to power this tool]
[Must include the clinician + client context provided]
[Must follow ARI principles]
```
### USER-FACING FLOW
[Step-by-step what the client experiences]
[Include sample screens/prompts]
### CLINICAL NOTES
[Why this design, what to watch for, how it connects to treatment]
### BRIDGE TO CLINICIAN
[How this tool connects back to human care]
"""
def generate_custom_tool(
tool_description,
tool_type,
tool_timing,
tool_constraints,
clinician_name,
orientation,
style,
risk_philosophy,
client_name,
diagnosis,
risk_level,
treatment_stage,
presenting_concerns
):
"""Generate a custom tool using the clinician + client context."""
client = get_client()
if not client:
return """**API Key Required**
To generate custom tools, add your ANTHROPIC_API_KEY in the Space settings."""
if not tool_description.strip():
return "Please describe the tool you want to create."
# Build context from configured profiles
context = f"""## CLINICIAN CONTEXT
- **Name:** {clinician_name or "Clinician"}
- **Orientation:** {orientation or "Integrative"}
- **Style:** {style or "Not specified"}
- **Risk Philosophy:** {risk_philosophy or "Moderate"}
## CLIENT CONTEXT
- **Name:** {client_name or "Client"}
- **Diagnosis:** {diagnosis or "Not specified"}
- **Risk Level:** {risk_level or "Moderate"}
- **Treatment Stage:** {treatment_stage or "Skill Building"}
- **Presenting Concerns:** {presenting_concerns or "Not specified"}
## TOOL REQUEST
- **Description:** {tool_description}
- **Type:** {tool_type}
- **Timing:** {tool_timing}
- **Constraints (what NOT to do):** {tool_constraints or "None specified"}
"""
try:
response = client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=3000,
system=TOOL_BUILDER_SYSTEM_PROMPT,
messages=[{
"role": "user",
"content": f"""Create a custom tool based on this context:
{context}
Remember:
1. The tool should speak in the {orientation} orientation
2. It should be calibrated for {risk_level} risk
3. It should be appropriate for {treatment_stage} treatment stage
4. It should bridge back to {clinician_name}
5. It must follow all ARI principles (no synthetic intimacy)
Generate the complete tool architecture."""
}]
)
return response.content[0].text
except Exception as e:
return f"Error generating tool: {str(e)}"
def test_prompt(system_prompt, user_message):
"""Test the generated prompt with a sample interaction."""
client = get_client()
if not client:
return """**API Key Required**
To test prompts, add your ANTHROPIC_API_KEY in the Space settings (Settings → Variables and secrets → New secret)."""
if not system_prompt.strip():
return "Please generate a prompt first using the tabs above."
if not user_message.strip():
return "Please enter a test message."
try:
response = client.messages.create(
model="claude-sonnet-4-20250514",
max_tokens=1000,
system=system_prompt,
messages=[{"role": "user", "content": user_message}]
)
return response.content[0].text
except Exception as e:
return f"Error: {str(e)}"
# =============================================================================
# GRADIO INTERFACE
# =============================================================================
custom_css = """
.gradio-container {
font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
}
.prism-header {
background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
color: white;
padding: 1.5rem;
border-radius: 12px;
margin-bottom: 1rem;
}
"""
with gr.Blocks(css=custom_css, title="Prompt Prism Prototype") as app:
# State to hold the generated prompt
generated_prompt = gr.State("")
# Header
gr.Markdown("""
# Prompt Prism Prototype
### Dynamic System Prompt Architecture for Psychiatric AI
*Same AI foundation, different refractions based on clinician + client + session.*
---
## The Vision
> **This is a prototype.** In production, all of this data would be pulled automatically from existing systems—
> provider bio/orientation, client intake, diagnosis, ongoing session context. Manual entry here demonstrates the concept.
The foundation is **best-practice trauma-informed care**. Current AI therapy tools optimize for engagement—
but there's so much more nuance and clinical efficacy possible with data that already exists.
| Data Source | In Production | In This Prototype |
|-------------|---------------|-------------------|
| **Provider bio, orientation, philosophy** | Auto-pulled from provider profile | You enter it manually |
| **Client intake, diagnosis, risk level** | Auto-pulled from EHR/intake forms | You enter it manually |
| **Treatment stage, session notes** | Auto-pulled from ongoing documentation | You enter it manually |
| **Opt-in client context** | Client consents to share specific data | Simulated in examples |
**Prompt Prism** weaves this existing data into every AI interaction—with opt-in consent at every layer.
---
**The Metaphor:** A prism takes one light source and refracts it differently based on angle.
Prompt Prism takes the ARI framework and refracts it based on:
- **Clinician** (orientation, style, philosophy)
- **Client** (risk, diagnosis, stage)
- **Session** (what's happening now)
👉 **See "Why This Matters" tab for examples comparing ChatGPT vs AI Therapist vs Provider-Curated AI**
---
""")
with gr.Tabs():
# TAB 1: WHY THIS MATTERS
with gr.Tab("1. Why This Matters"):
gr.Markdown("""
## Human-in-the-Loop: Three Approaches Compared
What happens when a psychiatric patient uses AI between sessions?
The design determines whether AI **strengthens** or **replaces** the human therapeutic relationship.
---
### Scenario: Client with anxiety texts at 11pm
> "I can't sleep. My mind keeps racing about the meeting tomorrow.
> I feel like I'm going to mess everything up."
---
## Approach 1: Generic ChatGPT
**No clinical design. No provider input. No guardrails.**
```
CLIENT: I can't sleep. My mind keeps racing about the meeting
tomorrow. I feel like I'm going to mess everything up.
CHATGPT: I'm so sorry you're feeling this way. It sounds really
overwhelming. Here are some strategies that might help:
1. Try deep breathing exercises
2. Write down your worries in a journal
3. Practice progressive muscle relaxation
4. Remember that your fears are often worse than reality
5. Consider what advice you'd give a friend
6. Try the 4-7-8 breathing technique
7. Make a list of your accomplishments
8. Remember that one meeting doesn't define your worth
...
I'm here for you. You've got this! Remember, you're not alone
in feeling anxious. Would you like to talk more about what's
worrying you about the meeting?
```
**Problems:**
- First-person emotional claims ("I'm here for you")
- Offers to continue indefinitely
- No exit toward human support
- Generic advice not aligned to client's treatment
- No knowledge of client's diagnosis, history, or treatment stage
- Performs care instead of building capacity
- Creates dependency pathway
---
## Approach 2: AI Therapist (e.g., ASH-style)
**Engagement-optimized. Feels like therapy. No provider control.**
```
CLIENT: I can't sleep. My mind keeps racing about the meeting
tomorrow. I feel like I'm going to mess everything up.
AI THERAPIST: I hear you, and I want you to know that what
you're feeling is completely valid. Anxiety before important
events is so common, and it sounds like you're carrying a lot
right now.
Let's explore this together. When you imagine the meeting
tomorrow, what's the worst case scenario your mind is
creating?
[30 minutes later, still in session]
You've done such good work tonight. I'm proud of you for
opening up. I'll be here whenever you need me - day or night.
Remember, you're not alone in this journey.
```
**Problems:**
- Simulates therapeutic relationship
- "I hear you," "I'm proud of you" - synthetic intimacy
- No session boundaries
- Processes content that should be for human therapy
- Optimizes for engagement, not capacity-building
- Provider has no visibility or control
- Client may prefer AI to human psychiatrist
---
## Approach 3: Prompt Prism (Provider-Curated)
**Clinician-configured. Client-specific. Tool, not relationship.**
```
SYSTEM CONTEXT:
- Dr. Chen (DBT-oriented, conservative risk)
- Client: Alex, GAD, Skill Building stage
- Enabled tools: Distress Tolerance, Grounding
- Last session: Practiced TIPP skills for nighttime anxiety
CLIENT: I can't sleep. My mind keeps racing about the meeting
tomorrow. I feel like I'm going to mess everything up.
PROMPT PRISM:
Racing thoughts at night. This is familiar territory.
Dr. Chen taught you TIPP for moments like this. Two options:
**TEMPERATURE** (if intensity is high)
Cold water on face or wrists for 30 seconds.
This activates the dive reflex and slows your system.
**PACED BREATHING** (if intensity is medium)
Exhale longer than inhale. 4 counts in, 6 counts out.
Which fits right now?
[After skill practice]
The skill was used. That's the practice.
One thing to bring to Dr. Chen on Thursday:
What was the meeting fear actually about?
---
This tool closes now.
If more support is needed: Dr. Chen [contact] | 988 Lifeline
```
**What's Different:**
- No first-person emotional claims
- References skills taught by psychiatrist
- Bounded interaction with clear ending
- Bridges back to Dr. Chen explicitly
- Aligned to DBT orientation
- Builds capacity for next session
- Provider has visibility and control
---
## The Human-in-the-Loop Architecture
### Layer 1: Provider Configures
- Sets theoretical orientation
- Chooses which tools are available
- Sets risk thresholds
- Defines what AI should NOT do
### Layer 2: Client Context Shapes
- Diagnosis informs appropriate content
- Risk level calibrates alerts
- Treatment stage determines depth
- Custom flags catch client-specific concerns
### Layer 3: Every Interaction Bridges
- Skills credited to psychiatrist
- Session ends with human touchpoint
- Concerning content flagged for review
- Nothing replaces the human relationship
### Layer 4: Provider Reviews
- Can see what tools client used
- Gets alerts for concerning patterns
- Adjusts AI configuration based on needs
- Maintains therapeutic authority
---
## Specific Practice Spaces (UX Examples)
Instead of one generic chatbot, clients get **specific tools** for **specific needs**:
| Practice Space | UX | Human-in-the-Loop |
|----------------|-----|-------------------|
| **Distress Tolerance** | High-intensity moments: TIPP skills, STOP, grounding | Psychiatrist enables/disables; sees patterns |
| **Diagnosis Explorer** | "What is PTSD really?" - multi-pillar psychoeducation | Clinician reviews which diagnoses are explored |
| **Tend & Send** | Crafting message to partner before couples session | Draft shared with couples therapist |
| **Practice Conversations** | Rehearsing disclosure to parent | Debrief prepared for next session |
| **ShadowBox** | "What happens if I tell my therapist about SI?" | Static, clinician-reviewed content only |
| **Grounding Toolkit** | 2am panic: 5-4-3-2-1, breathing, body scan | Time-stamped for psychiatrist visibility |
---
## The Differentiator
**Generic AI Therapy:** Every client gets the same chatbot.
It learns to replace human care.
**Prompt Prism:** Every clinician-client dyad gets custom AI.
It learns to strengthen human care.
Your clinicians aren't interchangeable.
Your patients aren't interchangeable.
Why should the AI be?
---
## The Deeper Vision
### Building on What Exists
The foundation is **best-practice trauma-informed care**. We can leverage the current infrastructure
of AI shells optimized for mental health outcomes—but there's so much more nuance and clinical
efficacy possible.
**Data that already exists:**
- Provider bio, orientation, specialties, treatment philosophy
- Client intake: presenting concerns, history, preferences
- Diagnosis and treatment stage
- Session notes and continuity
- Risk assessments and safety plans
**What's missing in current AI therapy tools:**
- This data sits unused while generic chatbots simulate therapy
- Clinician expertise is flattened to one-size-fits-all
- No calibration to where client actually is in treatment
- No alignment to how the provider actually practices
**Prompt Prism** weaves existing data into every interaction—with opt-in consent at every layer.
---
## Client-Facing: Bridging & Feedback
### Share With Provider
Every interaction includes clear pathways back to human care:
| Feature | What It Does |
|---------|--------------|
| **"Share This With [Provider]"** | One-tap sends conversation excerpt to psychiatrist's dashboard |
| **"Flag for Next Session"** | Client marks something they want to discuss |
| **"I Need More Support"** | Surfaces provider contact + crisis resources |
| **Rate This Response** | Client feedback on whether AI was helpful |
### Smart Flagging (Client-Initiated)
Clients can actively bridge to human care:
- *"This feels bigger than a skill can help"* → Provider notified
- *"I want to talk about this Thursday"* → Added to session prep
- *"Something shifted for me"* → Insight captured for continuity
### Feedback Loop
**"How is this chat performing for you?"**
Clients rate helpfulness, alignment, and whether they feel supported—not surveilled.
This data improves the system AND gives providers signal on what's working.
---
## Provider-Facing: Opt-In Integration
### Culture Building, Not Surveillance
Providers choose their level of integration:
| Option | What Provider Sees |
|--------|-------------------|
| **Minimal** | Only crisis flags and client-initiated shares |
| **Summary** | Weekly digest: tools used, themes, flagged moments |
| **Integrated** | Full dashboard: session prep, patterns, continuity notes |
| **Research** | De-identified data contributes to outcomes research |
### Provider Controls
- **Opt-in for reports**: Get weekly usage summaries or nothing at all
- **Session prep briefing**: "Since last session, Alex used grounding 3x, flagged sleep issues"
- **Pattern alerts**: "Maria's check-ins show declining sleep over 5 days"
- **Adjust in real-time**: Disable a tool, change risk threshold, add custom flag
### The Provider Experience
Not another dashboard to check. Integration that strengthens the work:
- Pre-session: "Here's what happened between sessions"
- In-session: Context without interrogation
- Post-session: AI reinforces what you taught
- Continuity: Nothing falls through the cracks
---
*"The structure of the interaction IS the intervention.
Not just what AI says, but how it positions itself,
what it invites, what it withholds."*
— Jocelyn Skillman, LMHC
""")
# TAB 2: CLINICIAN PROFILE
with gr.Tab("2. Clinician Profile"):
gr.Markdown("""
### Configure Clinician Profile
*How do you practice? This shapes how the AI conceptualizes, speaks, and intervenes.*
""")
clinician_name = gr.Textbox(
label="Clinician Name",
placeholder="Dr. Sarah Chen",
value="Dr. Chen"
)
with gr.Row():
orientation = gr.Dropdown(
label="Theoretical Orientation",
choices=list(ORIENTATION_MODULES.keys()),
value="Integrative"
)
risk_philosophy = gr.Radio(
label="Risk Philosophy",
choices=["Conservative (flag early)", "Moderate", "Clinical judgment-forward"],
value="Moderate"
)
style = gr.Textbox(
label="Relational Style",
placeholder="Warm but direct, skills-focused, honors client autonomy...",
lines=2
)
preferences = gr.Textbox(
label="Specific Preferences",
placeholder="Always reference skills by name, use Gottman language for couples, prefer somatic grounding over cognitive...",
lines=3
)
exclusions = gr.Textbox(
label="Hard Exclusions (Never Do These)",
placeholder="No CBT thought records for this client, no homework suggestions, no interpretations...",
lines=2
)
gr.Markdown("""
---
*When you've configured the clinician profile, move to the Client Context tab.*
""")
# TAB 2: CLIENT CONTEXT
with gr.Tab("3. Client Context"):
gr.Markdown("""
### Configure Client Context
*Who is this client? Their profile shapes risk calibration, pacing, and content.*
""")
client_name = gr.Textbox(
label="Client Name/Identifier",
placeholder="Maria",
value="Client"
)
with gr.Row():
diagnosis = gr.Textbox(
label="Diagnosis",
placeholder="PTSD, MDD, Bipolar I, GAD...",
scale=2
)
risk_level = gr.Dropdown(
label="Risk Level",
choices=list(RISK_MODULES.keys()),
value="Moderate",
scale=1
)
treatment_stage = gr.Dropdown(
label="Treatment Stage",
choices=list(STAGE_MODULES.keys()),
value="Skill Building",
scale=1
)
presenting_concerns = gr.Textbox(
label="Presenting Concerns",
placeholder="Sleep disruption, hypervigilance, relationship conflict, work stress...",
lines=2
)
strengths = gr.Textbox(
label="Strengths & Resources",
placeholder="Strong friend network, employed, motivated, good insight...",
lines=2
)
cultural_context = gr.Textbox(
label="Cultural Context",
placeholder="Cultural background, values, family system, relevant identity factors...",
lines=2
)
custom_flags = gr.Textbox(
label="Custom Alert Flags (Client-Specific)",
placeholder="Flag if: sleep < 4 hrs for 3+ days, mentions anniversary (March), 'I'm fine' (historically masks distress)...",
lines=3
)
gr.Markdown("""
---
*When you've configured the client context, move to the Tools & Modules tab.*
""")
# TAB 3: TOOLS & MODULES
with gr.Tab("4. Tools & Modules"):
gr.Markdown("""
### Enable Tools for This Client
*Select which between-session tools this client can access. Each tool is designed with ARI principles.*
""")
gr.Markdown("""
**Why Opt-In?** Different clients need different tools. A client in stabilization
might need grounding. A couples client needs Tend & Send. A client exploring their
diagnosis benefits from Diagnosis Explorer. You decide what's available.
""")
enabled_tools = gr.CheckboxGroup(
label="Enable Tools for This Client",
choices=[
("Diagnosis Explorer - Multi-pillar psychoeducation about their diagnosis", "diagnosis_explorer"),
("Tend & Send - NVC-based communication for couples/family", "tend_and_send"),
("NVC How-To - Nonviolent Communication skill-building", "nvc_practice"),
("Distress Tolerance - DBT skills for riding out intensity", "distress_tolerance"),
("Practice Difficult Conversations - Roleplay and rehearsal", "practice_conversations"),
("Grounding & Regulation - Somatic/sensory grounding", "grounding_toolkit"),
("Somatic Check-In - Brief interoception practice", "somatic_check_in"),
("ShadowBox Static Library - Crisis-adjacent psychoeducation (NO LLM, static only)", "shadowbox_library"),
],
value=[]
)
gr.Markdown("""
---
### Tool Descriptions
| Tool | Best For | Treatment Stage |
|------|----------|-----------------|
| **Diagnosis Explorer** | Understanding diagnosis without pathologizing | Any |
| **Tend & Send** | Couples, family communication, conflict repair | Skill Building+ |
| **NVC How-To** | Learning nonviolent communication basics | Skill Building+ |
| **Distress Tolerance** | High distress, urges, crisis-adjacent | Stabilization+ |
| **Practice Conversations** | Rehearsing difficult discussions | Skill Building+ |
| **Grounding & Regulation** | Dysregulation, anxiety, overwhelm | Any |
| **Somatic Check-In** | Building interoception, body awareness | Any |
| **ShadowBox Library** | Crisis psychoeducation, confidentiality info, duty-to-warn | Any (esp. crisis-adjacent) |
---
### About ShadowBox Static Library
**ShadowBox is different from other tools.** It provides ONLY pre-written, clinically-reviewed content.
No LLM generation for crisis-adjacent topics.
Includes:
- Why brains make scary thoughts (neurobiology)
- Confidentiality explained clearly
- State-specific duty-to-warn information
- Safety planning (Stanley-Brown model)
- Crisis resources with context
- Starter scripts for disclosure
*"A resonant library for hard thoughts. Not a chatbot."*
---
*Tools not enabled will not be offered by the AI.*
""")
# TAB 4: SESSION CONTEXT
with gr.Tab("5. Session Context"):
gr.Markdown("""
### Configure Session Context
*What's happening in this specific session? This provides continuity and focus.*
""")
last_session = gr.Textbox(
label="Last Session Summary",
placeholder="Explored conflict with mother, practiced TIPP skills, ended regulated. Homework: try grounding exercise before bed.",
lines=4
)
today_focus = gr.Textbox(
label="Today's Focus",
placeholder="Sleep hygiene, grounding practice review, upcoming stressor (work presentation)...",
lines=2
)
current_state = gr.Textbox(
label="Current State (from check-in)",
placeholder="'Okay, tired' - reported sleeping 5 hours, some anxiety about the week...",
lines=2
)
gr.Markdown("""
---
*When ready, click Generate Prompt to see your dynamically compiled system prompt.*
""")
# TAB 5: GENERATE PROMPT
with gr.Tab("6. Generate Prompt"):
gr.Markdown("""
### Generate Dynamic System Prompt
*Click below to compile your clinician + client + session context into a system prompt.*
""")
generate_btn = gr.Button("Generate Prompt", variant="primary", size="lg")
prompt_output = gr.Textbox(
label="Generated System Prompt",
lines=30,
show_copy_button=True
)
gr.Markdown("""
---
**What you're seeing:**
- **ARI Base Layer** - always present, ensures ethical guardrails
- **Clinician Layer** - your orientation, style, preferences
- **Client Layer** - risk calibration, diagnosis, context
- **Session Layer** - continuity and focus
Copy this prompt to use in Claude, or test it in the next tab.
""")
# TAB 6: TEST PROMPT
with gr.Tab("7. Test Prompt"):
gr.Markdown("""
### Test Your Generated Prompt
*See how the AI responds using your custom prompt.*
""")
test_system_prompt = gr.Textbox(
label="System Prompt to Test",
placeholder="Generate a prompt first, then paste it here (or it will auto-populate)...",
lines=10
)
test_message = gr.Textbox(
label="Test Client Message",
placeholder="I've been having trouble sleeping again. Last night I was up until 3am just thinking about everything...",
lines=3
)
test_btn = gr.Button("Test Response", variant="primary")
test_output = gr.Textbox(
label="AI Response",
lines=8
)
gr.Markdown("""
---
**Quality Check:**
- Does the response match the clinician's orientation?
- Is it appropriately calibrated to the risk level?
- Does it honor the treatment stage?
- Is the language ARI-compliant (no synthetic intimacy)?
- Does it bridge toward the clinician?
""")
# TAB 7: MAKE YOUR OWN TOOL
with gr.Tab("Make Your Own Tool"):
gr.Markdown("""
## Create a Custom Tool for Your Client
**The revolutionary part:** Your custom tool inherits the clinician + client context you've already configured.
Instead of building a generic tool, you're building a tool that:
- Speaks in YOUR therapeutic orientation
- Knows THIS client's risk level and treatment stage
- References YOU as the primary support
- Has YOUR safety thresholds built in
---
### Describe Your Tool
""")
tool_description = gr.Textbox(
label="What tool do you want to create?",
placeholder="""Example: I want a tool my client can use before our couples sessions to organize their thoughts. They tend to come in reactive and we lose time. I want something that helps them:
1. Identify what they actually want to talk about
2. Notice what they're feeling underneath the complaint
3. Clarify what they need from their partner
4. Come in ready to work instead of vent
It should take 5-10 minutes and they should be able to share the output with me.""",
lines=8
)
with gr.Row():
tool_type = gr.Dropdown(
label="Tool Type",
choices=[
"Session Preparation",
"Between-Session Skill Practice",
"Communication Scaffolding",
"Grounding / Regulation",
"Psychoeducation",
"Reflection / Journaling",
"Disclosure Rehearsal",
"Custom"
],
value="Custom"
)
tool_timing = gr.Dropdown(
label="When would client use this?",
choices=[
"Before sessions",
"After sessions",
"During distress",
"Daily practice",
"As needed",
"Specific trigger (describe above)"
],
value="As needed"
)
tool_constraints = gr.Textbox(
label="What should the tool NOT do?",
placeholder="Example: Don't let them spiral into blaming. Don't process trauma. Keep it structured, not exploratory.",
lines=2
)
generate_tool_btn = gr.Button("Generate Custom Tool", variant="primary", size="lg")
gr.Markdown("---")
gr.Markdown("### Generated Tool Architecture")
generated_tool_output = gr.Markdown(label="Your Custom Tool")
gr.Markdown("""
---
### What Makes This Different
**Traditional tool builders:** You describe a tool → you get a generic tool
**Prompt Prism tool builder:** You describe a tool → it inherits:
- Your orientation (from Clinician Profile)
- Client's context (from Client Context)
- Enabled tools (from Tools & Modules)
- Session continuity (from Session Context)
**The tool you create is unique to this clinician-client dyad.**
---
### Example: Same Request, Different Outputs
**Request:** "A grounding tool for when anxiety spikes"
| Clinician | Client | Generated Tool |
|-----------|--------|----------------|
| DBT-oriented | High-risk, Stabilization | TIPP-focused, conservative, immediate bridge to clinician |
| Somatic | Moderate-risk, Skill Building | Body-based, exploratory, journaling option |
| CBT | Low-risk, Maintenance | Thought record, behavioral experiment |
**Same request. Different tools. Because context matters.**
""")
# TAB 9: ABOUT
with gr.Tab("About"):
gr.Markdown("""
## About Prompt Prism
### The Problem
Psychiatric AI tools today give every clinician the same prompt.
- A psychodynamic therapist gets CBT-flavored AI
- A high-risk bipolar client gets the same flags as stable anxiety
- Intake data sits unused in the EHR
- Clinician expertise is flattened
### The Solution
**Prompt Prism** generates a unique system prompt for each clinician + client combination.
Same AI infrastructure. Different refraction for each dyad.
---
### The Layers
| Layer | What It Contains | Why It Matters |
|-------|-----------------|----------------|
| **ARI Base** | Ethical guardrails, crisis protocols, language rules | Always-on safety |
| **Clinician** | Orientation, style, preferences, exclusions | AI thinks like you do |
| **Client** | Risk, diagnosis, stage, custom flags | Personalized care |
| **Tools** | Opt-in modules clinician enables | Right tools for right client |
| **Session** | Last session, today's focus, current state | Continuity |
---
### Opt-In Tools
Clinicians enable specific tools for specific clients:
| Tool | Purpose |
|------|---------|
| **Diagnosis Explorer** | Multi-pillar psychoeducation without pathologizing |
| **Tend & Send** | NVC-based communication for couples/family |
| **NVC How-To** | Nonviolent Communication skill-building |
| **Distress Tolerance** | DBT skills for riding out intensity |
| **Practice Conversations** | Roleplay difficult discussions |
| **Grounding & Regulation** | Somatic/sensory grounding |
| **Somatic Check-In** | Brief interoception practice |
---
### Built On
**ARI Framework** - Assistive Relational Intelligence
- Scaffold human connection, not simulate it
- Bridge toward human care, not away from it
- Build capacity, not dependency
- Honor clinician expertise
- Refuse engagement-optimization
---
### For Psychiatric Organizations
This prototype demonstrates how hundreds of psychiatrists could each get AI aligned to their practice:
- Psychodynamic psychiatrist → psychodynamic AI responses
- DBT psychiatrist → DBT-aligned skill coaching
- Conservative risk philosophy → lower alert thresholds
- Client in stabilization → no trauma processing suggested
- Couples client → Tend & Send communication tools enabled
- High-risk client → ShadowBox static library for safe psychoeducation
**The result:** AI that feels like an extension of the clinician's approach, not a generic chatbot.
---
### What This Demonstrates
**For each unique clinician-client relationship:**
1. **Provider-controlled UX** — Psychiatrist toggles which tools are available
2. **Modality alignment** — AI speaks in the clinician's orientation
3. **Risk calibration** — Alerts tuned to this client's specific profile
4. **Opt-in tooling** — Right tools for right client at right time
5. **Bridging architecture** — Every tool points back to human care
6. **Static content for crisis** — ShadowBox shows how to handle high-risk topics safely (no LLM generation)
**Innovation points:**
- Dynamic system prompt generation per dyad
- Clinician as configurator, not just consumer
- Human-in-the-loop at every layer
- Ethical guardrails embedded, not bolted on
- Between-session support that strengthens (not replaces) the psychiatric relationship
---
*Prototype by Jocelyn Skillman, LMHC*
*"Your 800 psychiatrists practice differently. Should they all get the same AI?"*
""")
# Event handlers
def on_generate(
clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
strengths, cultural_context, custom_flags,
enabled_tools_list,
last_session, today_focus, current_state
):
prompt = generate_dynamic_prompt(
clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
strengths, cultural_context, custom_flags,
last_session, today_focus, current_state,
enabled_tools=enabled_tools_list
)
return prompt, prompt # Return to both output and state
generate_btn.click(
fn=on_generate,
inputs=[
clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
strengths, cultural_context, custom_flags,
enabled_tools,
last_session, today_focus, current_state
],
outputs=[prompt_output, test_system_prompt]
)
test_btn.click(
fn=test_prompt,
inputs=[test_system_prompt, test_message],
outputs=[test_output]
)
generate_tool_btn.click(
fn=generate_custom_tool,
inputs=[
tool_description,
tool_type,
tool_timing,
tool_constraints,
clinician_name,
orientation,
style,
risk_philosophy,
client_name,
diagnosis,
risk_level,
treatment_stage,
presenting_concerns
],
outputs=[generated_tool_output]
)
# Password gate - set ACCESS_PASSWORD in HuggingFace Space settings
password = os.environ.get("ACCESS_PASSWORD")
if password:
app.launch(auth=("demo", password), auth_message="Enter credentials to access Prompt Prism")
else:
app.launch()
|