v1.2.0 — Empirical tier restructure, human-proof agentic challenges

Major changes since v0.8.0:

Tier restructure (empirically calibrated):
  Only types where GPT-5.2 achieves 100% are active.
  Easy (4):    simple_math, string_math, binary, pattern
  Medium (2):  sorting, word_math
  Hard (3):    nested_operations, base_conversion_chain, power_mod [NEW]
  Agentic (1): chained_arithmetic [NEW]
  Shelved (17): All character-manipulation types reserved for future models

New challenge types:
  - chained_arithmetic: 4-step chain with 4 operation patterns
    GPT-5.2: 100% | GPT-4o: 30% | Humans: 15-20s without paper
  - power_mod: base^exp mod m
    GPT-5.2: 100% | GPT-4o: 80%

Dynamic mode:
  - Generates multi-step arithmetic (no character manipulation)
  - Iteratively refined through 4 rounds against real API
  - GPT-5.2 verified at 100% solve rate

Frontend (challenge.llm.kaveenk.com):
  - Human/Agent toggle (agent mode hides everything, shows SKILL.md)
  - All tiers displayed with empirical percentages
  - GPT-5.2 row in model compatibility table
  - Shelved types section (greyed out)
  - Lock mode TTL: 5 seconds
  - Toggle centered properly

Tests: 219 unit + 23 E2E, 7/7 CI green
GPT-5.2 calibration data: tests/calibration_gpt52.json

Author: root

package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-challenge",
-  "version": "1.1.0",
+  "version": "1.2.0",
   "description": "Drop-in LLM authentication for any API endpoint. Reasoning puzzles that agents solve once, then pass through forever. Stateless HMAC tokens, no database.",
   "main": "src/agentchallenge.js",
   "type": "module",

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "agent-challenge"
-version = "1.1.0"
+version = "1.2.0"
 description = "LLM-solvable challenge-response authentication for AI agent APIs"
 readme = "README.md"
 license = {text = "MIT"}

src/agentchallenge/__init__.py

@@ -34,7 +34,7 @@
 
 from .safe_solve import validate_prompt, safe_solve, ISOLATION_PROMPT
 
-__version__ = "1.1.0"
+__version__ = "1.2.0"
 __all__ = [
     "AgentChallenge", "Challenge", "VerifyResult", "GateResult",
     "CHALLENGE_TYPES", "validate_prompt", "safe_solve", "ISOLATION_PROMPT",