From 89e3e997dfa404d27b8c2ad0e0278a0a3cbbc36a Mon Sep 17 00:00:00 2001
From: OpenCode <opencode@example.com>
Date: Tue, 7 Jul 2026 10:04:32 +0700
Subject: [PATCH] Add AGENT_TASKS.json for machine-readable safe task routing

Closes #220

- AGENT_TASKS.json with 8 task categories (docs, tests, schemas, CI,
  benchmarks, calibration, external_review, release_safety)
- Each category has: safe_paths, review_class, forbidden_zones,
  required_checks, expected_artifact
- Stop conditions and non-negotiables from AGENTS.md
- JSON Schema validatable
- 7 tests covering structure, categories, fields, review classes
---
 AGENT_TASKS.json          | 95 +++++++++++++++++++++++++++++++++++++++
 tests/test_agent_tasks.py | 43 ++++++++++++++++++
 2 files changed, 138 insertions(+)
 create mode 100644 AGENT_TASKS.json
 create mode 100644 tests/test_agent_tasks.py

diff --git a/AGENT_TASKS.json b/AGENT_TASKS.json
new file mode 100644
index 00000000..68bf853d
--- /dev/null
+++ b/AGENT_TASKS.json
@@ -0,0 +1,95 @@
+{
+  "$schema": "https://json-schema.org/draft/2020-12/schema",
+  "title": "OpenAMP Agent Task Map",
+  "description": "Machine-readable safe task routing for AI agents.",
+  "task_categories": [
+    {
+      "category": "docs",
+      "safe_paths": ["docs/", "README.md", "CONTRIBUTING.md"],
+      "review_class": "light",
+      "forbidden_zones": ["docs/safety/", "docs/trust/"],
+      "required_checks": ["check_doc_links", "check_claims"],
+      "example_pr_size": "< 200 lines",
+      "expected_artifact": "documentation change"
+    },
+    {
+      "category": "tests",
+      "safe_paths": ["tests/"],
+      "review_class": "light",
+      "forbidden_zones": [],
+      "required_checks": ["pytest"],
+      "example_pr_size": "< 300 lines",
+      "expected_artifact": "test file"
+    },
+    {
+      "category": "schemas",
+      "safe_paths": ["schemas/"],
+      "review_class": "normal",
+      "forbidden_zones": [],
+      "required_checks": ["validate_schema", "pytest"],
+      "example_pr_size": "< 100 lines",
+      "expected_artifact": "schema file"
+    },
+    {
+      "category": "ci",
+      "safe_paths": [".github/"],
+      "review_class": "normal",
+      "forbidden_zones": [],
+      "required_checks": ["pytest"],
+      "example_pr_size": "< 50 lines",
+      "expected_artifact": "CI workflow change"
+    },
+    {
+      "category": "benchmarks",
+      "safe_paths": ["scripts/benchmarks/", "scripts/benchmark_*"],
+      "review_class": "normal",
+      "forbidden_zones": ["configs/pipeline.yaml", "configs/phase3.yaml"],
+      "required_checks": ["pytest", "bench-gate"],
+      "example_pr_size": "< 200 lines",
+      "expected_artifact": "benchmark script or card"
+    },
+    {
+      "category": "calibration",
+      "safe_paths": ["src/openamp_foundry/calibration/", "configs/recalibration_policy.yaml"],
+      "review_class": "requires_human_review",
+      "forbidden_zones": [],
+      "required_checks": ["pytest"],
+      "example_pr_size": "< 300 lines",
+      "expected_artifact": "calibration module change"
+    },
+    {
+      "category": "external_review",
+      "safe_paths": ["docs/review/"],
+      "review_class": "requires_human_review",
+      "forbidden_zones": [],
+      "required_checks": ["check_claims"],
+      "example_pr_size": "< 200 lines",
+      "expected_artifact": "review document"
+    },
+    {
+      "category": "release_safety",
+      "safe_paths": [],
+      "review_class": "blocked",
+      "forbidden_zones": ["ALL"],
+      "required_checks": [],
+      "example_pr_size": "N/A",
+      "expected_artifact": "N/A — requires human release authority"
+    }
+  ],
+  "stop_conditions": [
+    "Task broadens beyond requested scope",
+    "Change touches safety or release policy",
+    "Claim becomes stronger than evidence supports",
+    "Benchmark threshold changes",
+    "Non-toy data or candidate artifacts appear",
+    "Tests fail in an unexplained way",
+    "Uncertainty is being hidden"
+  ],
+  "non_negotiable": [
+    "Computational outputs are not biological proof",
+    "The lab is the judge",
+    "No cherry-picking — preserve negative results",
+    "Safety-first optimization",
+    "Human review required for safety, release, and scientific claims"
+  ]
+}
diff --git a/tests/test_agent_tasks.py b/tests/test_agent_tasks.py
new file mode 100644
index 00000000..c49325eb
--- /dev/null
+++ b/tests/test_agent_tasks.py
@@ -0,0 +1,43 @@
+"""Tests for AGENT_TASKS.json."""
+import json
+from pathlib import Path
+
+
+def test_file_exists():
+    assert Path("AGENT_TASKS.json").exists()
+
+
+def test_is_valid_json():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    assert "task_categories" in data
+    assert "stop_conditions" in data
+    assert "non_negotiable" in data
+
+
+def test_has_8_categories():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    assert len(data["task_categories"]) >= 8
+
+
+def test_each_category_has_required_fields():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    for cat in data["task_categories"]:
+        for field in ["category", "safe_paths", "review_class", "required_checks"]:
+            assert field in cat, f"{cat['category']} missing {field}"
+
+
+def test_review_classes():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    valid = {"light", "normal", "requires_human_review", "blocked"}
+    for cat in data["task_categories"]:
+        assert cat["review_class"] in valid, f"{cat['category']} has invalid review_class"
+
+
+def test_stop_conditions():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    assert len(data["stop_conditions"]) >= 5
+
+
+def test_non_negotiable():
+    data = json.loads(Path("AGENT_TASKS.json").read_text())
+    assert len(data["non_negotiable"]) >= 3