diff --git a/apps/chat/routes.py b/apps/chat/routes.py
index 21b1bb53b..f45731dec 100644
--- a/apps/chat/routes.py
+++ b/apps/chat/routes.py
@@ -6,6 +6,7 @@ from __future__ import annotations
 import logging
 import os
 import time
+from pathlib import Path
 from typing import Any
 
 from flask import Blueprint, jsonify, render_template, request
@@ -93,13 +94,6 @@ def index():
     return render_template("app.html")
 
 
-TITLE_SYSTEM_INSTRUCTION = (
-    "Take the user provided text and come up with a three word title that "
-    "concisely but uniquely identifies the user's request for quick reference "
-    "and recall. Output only the three word title, nothing else."
-)
-
-
 def _check_provider_api_key(provider: str) -> str | None:
     """Check if provider API key is set and return error message if not.
 
@@ -119,11 +113,14 @@ def _check_provider_api_key(provider: str) -> str | None:
 
 def generate_chat_title(message: str) -> str:
     """Generate a short title for a chat message using configured provider."""
+    from think.utils import load_prompt
+
+    prompt = load_prompt("title", base_dir=Path(__file__).parent)
     try:
         title = generate(
             contents=message,
             context="app.chat.title",
-            system_instruction=TITLE_SYSTEM_INSTRUCTION,
+            system_instruction=prompt.text,
             max_output_tokens=50,
             timeout_s=10,
         ).strip()
diff --git a/apps/chat/title.md b/apps/chat/title.md
new file mode 100644
index 000000000..9cb08a4f8
--- /dev/null
+++ b/apps/chat/title.md
@@ -0,0 +1,7 @@
+---
+context: app.chat.title
+tier: 3
+label: Chat Title Generation
+group: Apps
+---
+Take the user provided text and come up with a three word title that concisely but uniquely identifies the user's request for quick reference and recall. Output only the three word title, nothing else.
diff --git a/docs/PROVIDERS.md b/docs/PROVIDERS.md
index 7ddbbb3d0..5940bbd35 100644
--- a/docs/PROVIDERS.md
+++ b/docs/PROVIDERS.md
@@ -249,19 +249,18 @@ Context strings determine provider and model selection. Providers receive alread
 - Other contexts: `{module}.{feature}[.{operation}]`
   - Examples: `observe.describe.frame`, `app.chat.title`
 
-**Dynamic discovery:** Categories and muse configs express their own tier/label/group in their configs:
-- Categories: `observe/categories/*.json` - add `tier`, `label`, `group` fields
+**Dynamic discovery:** All context metadata (tier/label/group) is defined in prompt .md files via YAML frontmatter:
+- Prompt files: Listed in `PROMPT_PATHS` in `think/models.py` - add `context`, `tier`, `label`, `group` fields
+- Categories: `observe/categories/*.md` - add `tier`, `label`, `group` fields
 - System muse: `muse/*.md` - add `tier`, `label`, `group` fields in frontmatter
 - App muse: `apps/*/muse/*.md` - add `tier`, `label`, `group` fields in frontmatter
 
-These are discovered at runtime and merged with static defaults. Use `get_context_registry()` to get the complete context map including discovered entries.
-
-See `CONTEXT_DEFAULTS` in `think/models.py` for static context patterns (non-discoverable contexts like `observe.detect.*`).
+All contexts are discovered at runtime. Use `get_context_registry()` to get the complete context map.
 
 **Resolution** (handled by `think/models.py` `resolve_provider()`):
 1. Exact match in journal.json `providers.contexts`
 2. Glob pattern match (fnmatch) with specificity ranking
-3. Dynamic context registry (static defaults + discovered categories/agents)
+3. Dynamic context registry (discovered prompts, categories, muse configs)
 4. Default provider/tier from config
 
 Providers don't implement routing - they receive the resolved model.
diff --git a/observe/describe.md b/observe/describe.md
index 20022085a..8cae7409d 100644
--- a/observe/describe.md
+++ b/observe/describe.md
@@ -1,3 +1,9 @@
+---
+context: observe.describe.frame
+tier: 3
+label: Screen Categorization
+group: Observe
+---
 You have one job: identify the primary foreground and (if present) secondary app categories in this desktop screenshot, and return ONLY this JSON:
 
 {
diff --git a/observe/describe.py b/observe/describe.py
index 01d69e117..e7bcd0d7a 100644
--- a/observe/describe.py
+++ b/observe/describe.py
@@ -420,7 +420,7 @@ class VideoProcessor:
                 output_file.write(json.dumps(metadata) + "\n")
                 output_file.flush()
 
-            # Resolve model for frame description (tier comes from CONTEXT_DEFAULTS)
+            # Resolve model for frame description (tier from describe.md frontmatter)
             _, frame_model = resolve_provider("observe.describe.frame")
 
             # Create vision requests for all qualified frames
diff --git a/observe/enrich.md b/observe/enrich.md
index c825db963..76d53a2c7 100644
--- a/observe/enrich.md
+++ b/observe/enrich.md
@@ -1,3 +1,9 @@
+---
+context: observe.enrich
+tier: 2
+label: Audio Enrichment
+group: Observe
+---
 You are correcting and enriching an audio transcript. You receive numbered statements with transcribed text and corresponding audio clips.
 
 For each statement:
diff --git a/observe/enrich.py b/observe/enrich.py
index c72a0f89c..2563359dd 100644
--- a/observe/enrich.py
+++ b/observe/enrich.py
@@ -101,7 +101,7 @@ def enrich_transcript(
                 types.Part.from_bytes(data=audio_bytes, mime_type="audio/flac")
             )
 
-        # Call LLM (tier defaults to LITE via CONTEXT_DEFAULTS)
+        # Call LLM (tier from enrich.md frontmatter)
         logger.info(f"Enriching {len(statements)} statements...")
         t0 = time.perf_counter()
 
diff --git a/observe/extract.md b/observe/extract.md
index ba27731d2..c0bd0071e 100644
--- a/observe/extract.md
+++ b/observe/extract.md
@@ -1,3 +1,9 @@
+---
+context: observe.extract.selection
+tier: 2
+label: Frame Selection
+group: Observe
+---
 You are analyzing frame categorizations from a desktop screencast recording to select frames for detailed content extraction.
 
 Given a time-series of frames with categories and visual descriptions, select the frames most valuable for text extraction and content analysis. Aim to select around $max_extractions frames total, fewer if the content is repetitive.
diff --git a/observe/transcribe/gemini.md b/observe/transcribe/gemini.md
index bd7e3c322..23741967e 100644
--- a/observe/transcribe/gemini.md
+++ b/observe/transcribe/gemini.md
@@ -1,3 +1,9 @@
+---
+context: observe.transcribe.gemini
+tier: 2
+label: Audio Transcription (Gemini)
+group: Observe
+---
 You are accurately transcribing audio and indentifying distinct. Return a JSON object with individual speech segments that represent separate statements or sentences.
 
 ## Output Format
diff --git a/tests/test_models.py b/tests/test_models.py
index 02058d399..ff92c6433 100644
--- a/tests/test_models.py
+++ b/tests/test_models.py
@@ -9,7 +9,6 @@ from think.models import (
     CLAUDE_HAIKU_4,
     CLAUDE_OPUS_4,
     CLAUDE_SONNET_4,
-    CONTEXT_DEFAULTS,
     DEFAULT_PROVIDER,
     DEFAULT_TIER,
     GEMINI_FLASH,
@@ -18,6 +17,7 @@ from think.models import (
     GPT_5,
     GPT_5_MINI,
     GPT_5_NANO,
+    PROMPT_PATHS,
     PROVIDER_DEFAULTS,
     TIER_FLASH,
     TIER_LITE,
@@ -196,41 +196,56 @@ def test_tier_constants():
     assert DEFAULT_PROVIDER == "google"
 
 
-def test_context_defaults_structure():
-    """Test CONTEXT_DEFAULTS has required fields for each entry."""
-    required_keys = {"tier", "label", "group"}
+def test_prompt_paths_exist():
+    """Test all PROMPT_PATHS files exist and have valid frontmatter."""
+    from pathlib import Path
+
+    import frontmatter
+
+    base_dir = Path(__file__).parent.parent  # Project root
+    required_keys = {"context", "tier", "label", "group"}
+
+    for rel_path in PROMPT_PATHS:
+        path = base_dir / rel_path
+        assert path.exists(), f"Prompt file not found: {rel_path}"
+
+        post = frontmatter.load(path)
+        meta = post.metadata or {}
 
-    for context, config in CONTEXT_DEFAULTS.items():
-        assert isinstance(config, dict), f"{context} should be a dict"
         assert required_keys <= set(
-            config.keys()
-        ), f"{context} missing keys: {required_keys - set(config.keys())}"
-        assert config["tier"] in (
+            meta.keys()
+        ), f"{rel_path} missing keys: {required_keys - set(meta.keys())}"
+        assert meta["tier"] in (
             TIER_PRO,
             TIER_FLASH,
             TIER_LITE,
-        ), f"{context} has invalid tier: {config['tier']}"
+        ), f"{rel_path} has invalid tier: {meta['tier']}"
         assert (
-            isinstance(config["label"], str) and config["label"]
-        ), f"{context} has invalid label: {config['label']}"
+            isinstance(meta["label"], str) and meta["label"]
+        ), f"{rel_path} has invalid label: {meta['label']}"
         assert (
-            isinstance(config["group"], str) and config["group"]
-        ), f"{context} has invalid group: {config['group']}"
+            isinstance(meta["group"], str) and meta["group"]
+        ), f"{rel_path} has invalid group: {meta['group']}"
+
+
+def test_prompt_contexts_in_registry():
+    """Test prompt contexts are discovered and in registry."""
+    registry = get_context_registry()
 
+    # Verify known prompt contexts exist with correct values
+    assert "observe.describe.frame" in registry
+    assert registry["observe.describe.frame"]["tier"] == TIER_LITE
+    assert registry["observe.describe.frame"]["group"] == "Observe"
 
-def test_context_defaults_known_entries():
-    """Test specific known CONTEXT_DEFAULTS entries."""
-    # Verify some known entries exist with correct values
-    assert "observe.describe.frame" in CONTEXT_DEFAULTS
-    assert CONTEXT_DEFAULTS["observe.describe.frame"]["tier"] == TIER_LITE
-    assert CONTEXT_DEFAULTS["observe.describe.frame"]["group"] == "Observe"
+    assert "app.chat.title" in registry
+    assert registry["app.chat.title"]["tier"] == TIER_LITE
+    assert registry["app.chat.title"]["group"] == "Apps"
 
-    # agent.* static defaults removed - now discovered from muse/*.md
-    assert "agent.*" not in CONTEXT_DEFAULTS
+    assert "observe.enrich" in registry
+    assert registry["observe.enrich"]["tier"] == TIER_FLASH
 
-    assert "app.chat.title" in CONTEXT_DEFAULTS
-    assert CONTEXT_DEFAULTS["app.chat.title"]["tier"] == TIER_LITE
-    assert CONTEXT_DEFAULTS["app.chat.title"]["group"] == "Apps"
+    assert "detect.created" in registry
+    assert registry["detect.created"]["tier"] == TIER_LITE
 
 
 def test_provider_defaults_structure():
@@ -362,30 +377,38 @@ def test_resolve_provider_invalid_tier(use_fixtures_journal, monkeypatch, tmp_pa
 # ---------------------------------------------------------------------------
 
 
-def test_context_registry_includes_static_defaults():
-    """Test that registry includes all static CONTEXT_DEFAULTS entries."""
+def test_context_registry_includes_prompt_contexts():
+    """Test that registry includes all contexts from PROMPT_PATHS."""
+    from pathlib import Path
+
+    import frontmatter
+
     registry = get_context_registry()
+    base_dir = Path(__file__).parent.parent
+
+    # All prompt contexts should be in registry with correct tier
+    for rel_path in PROMPT_PATHS:
+        path = base_dir / rel_path
+        post = frontmatter.load(path)
+        meta = post.metadata or {}
+        context = meta.get("context")
 
-    # All static defaults should be in registry
-    for context in CONTEXT_DEFAULTS:
-        assert context in registry, f"Static default {context} not in registry"
-        assert registry[context]["tier"] == CONTEXT_DEFAULTS[context]["tier"]
+        assert context in registry, f"Prompt context {context} not in registry"
+        assert registry[context]["tier"] == meta["tier"]
 
 
 def test_context_registry_includes_categories():
     """Test that registry includes discovered category contexts."""
     registry = get_context_registry()
 
-    # Should have category entries (from observe/categories/*.json)
+    # Should have category entries (from observe/categories/*.md)
     category_contexts = [k for k in registry if k.startswith("observe.describe.")]
 
-    # Should have more than just the static entries (frame and wildcard)
-    assert len(category_contexts) > 2, "Should discover category contexts"
+    # Should have frame + all categories (browsing, code, gaming, etc.)
+    assert len(category_contexts) > 5, "Should discover category contexts"
 
     # Each category context should have required fields
     for context in category_contexts:
-        if context == "observe.describe.*":
-            continue  # Skip wildcard
         assert "tier" in registry[context]
         assert "label" in registry[context]
         assert "group" in registry[context]
diff --git a/think/detect_created.md b/think/detect_created.md
index cb30470aa..a4a8cb6fe 100644
--- a/think/detect_created.md
+++ b/think/detect_created.md
@@ -1,3 +1,9 @@
+---
+context: detect.created
+tier: 3
+label: Date Detection
+group: Import
+---
 You are an expert at analyzing media file metadata to determine creation timestamps. Analyze the provided exiftool metadata output and extract the most accurate creation time.
 
 Guidelines:
diff --git a/think/detect_transcript_json.md b/think/detect_transcript_json.md
index f706fec8a..8bc7fa48f 100644
--- a/think/detect_transcript_json.md
+++ b/think/detect_transcript_json.md
@@ -1,3 +1,9 @@
+---
+context: observe.detect.json
+tier: 2
+label: Normalization
+group: Import
+---
 You are a transcript processing assistant. Convert the provided transcript segment into structured JSON format.
 
 ## Input Format:
diff --git a/think/detect_transcript_segment.md b/think/detect_transcript_segment.md
index d8c25d2d4..7893e6585 100644
--- a/think/detect_transcript_segment.md
+++ b/think/detect_transcript_segment.md
@@ -1,3 +1,9 @@
+---
+context: observe.detect.segment
+tier: 2
+label: Segmentation
+group: Import
+---
 You are a transcript analyzer that identifies 5-minute segment boundaries with absolute timestamps.
 
 TASK: Find ~5-minute segment boundaries and return their line numbers with absolute time-of-day timestamps.
diff --git a/think/models.py b/think/models.py
index 10d056e3d..a8a84c22b 100644
--- a/think/models.py
+++ b/think/models.py
@@ -92,15 +92,10 @@ class IncompleteJSONError(ValueError):
 
 
 # ---------------------------------------------------------------------------
-# Context defaults: context pattern -> {tier, label, group}
+# Prompt context discovery
 #
-# These define the default tier for each context when not overridden in config.
-# Patterns support glob-style matching (fnmatch).
-#
-# Each entry contains:
-#   - tier: Default tier (TIER_PRO, TIER_FLASH, TIER_LITE)
-#   - label: Human-readable name for settings UI
-#   - group: Category for grouping in settings UI
+# Context metadata (tier, label, group) is defined in prompt .md files via
+# YAML frontmatter. This eliminates duplication between code and config.
 #
 # NAMING CONVENTION:
 #   {module}.{feature}[.{operation}]
@@ -112,82 +107,31 @@ class IncompleteJSONError(ValueError):
 #   - muse.entities.observer    -> muse module, entities app, observer config
 #   - app.chat.title            -> apps module, chat app, title operation
 #
-# DYNAMIC DISCOVERY:
-#   Categories (observe/categories/*.json) and agents (muse/*.md,
-#   apps/*/muse/*.md) can express tier/label/group in their frontmatter.
-#   These are discovered at runtime and merged with the static defaults below.
+# DISCOVERY SOURCES:
+#   1. Prompt files listed in PROMPT_PATHS (with context in frontmatter)
+#   2. Categories from observe/categories/*.md (tier/label/group in frontmatter)
+#   3. Muse configs from muse/*.md and apps/*/muse/*.md
 #
 # When adding new contexts:
-#   1. Use module prefix matching the package (observe, think, app)
-#   2. Add specific operations as suffixes when granular control is needed
-#   3. Use wildcards sparingly - prefer explicit entries for clarity
-#   4. If not listed here, context falls back to DEFAULT_TIER (FLASH)
-#   5. For categories/agents, prefer adding tier/label/group to JSON configs
+#   1. Create a .md prompt file with YAML frontmatter containing:
+#      context, tier, label, group
+#   2. Add the path to PROMPT_PATHS
+#   3. If not listed, context falls back to DEFAULT_TIER (FLASH)
 # ---------------------------------------------------------------------------
 
-# Static context defaults - non-discoverable contexts only
-# Categories and agents express their own tier/label/group in JSON configs
-CONTEXT_DEFAULTS: Dict[str, Dict[str, Any]] = {
-    # Observe pipeline - screen and audio capture processing
-    "observe.describe.frame": {
-        "tier": TIER_LITE,
-        "label": "Screen Categorization",
-        "group": "Observe",
-    },
-    # Fallback for categories without explicit tier in their JSON
-    "observe.describe.*": {
-        "tier": TIER_FLASH,
-        "label": "Screen Extraction",
-        "group": "Observe",
-    },
-    "observe.detect.segment": {
-        "tier": TIER_FLASH,
-        "label": "Segmentation",
-        "group": "Import",
-    },
-    "observe.detect.json": {
-        "tier": TIER_FLASH,
-        "label": "Normalization",
-        "group": "Import",
-    },
-    "observe.enrich": {
-        "tier": TIER_FLASH,
-        "label": "Audio Enrichment",
-        "group": "Observe",
-    },
-    "observe.transcribe.gemini": {
-        "tier": TIER_FLASH,
-        "label": "Audio Transcription (Gemini)",
-        "group": "Observe",
-    },
-    "observe.extract.selection": {
-        "tier": TIER_FLASH,
-        "label": "Frame Selection",
-        "group": "Observe",
-    },
-    "observe.summarize": {
-        "tier": TIER_FLASH,
-        "label": "Summarization",
-        "group": "Import",
-    },
-    # Utilities - miscellaneous processing tasks
-    "detect.created": {
-        "tier": TIER_LITE,
-        "label": "Date Detection",
-        "group": "Import",
-    },
-    "planner.generate": {
-        "tier": TIER_FLASH,
-        "label": "Agent Prompt Generation",
-        "group": "Think",
-    },
-    # Apps - application-specific contexts
-    "app.chat.title": {
-        "tier": TIER_LITE,
-        "label": "Chat Title Generation",
-        "group": "Apps",
-    },
-}
+# Flat list of prompt files that define context metadata in frontmatter.
+# Each must have: context, tier, label, group in YAML frontmatter.
+PROMPT_PATHS: List[str] = [
+    "observe/describe.md",
+    "observe/enrich.md",
+    "observe/extract.md",
+    "observe/transcribe/gemini.md",
+    "think/detect_created.md",
+    "think/detect_transcript_segment.md",
+    "think/detect_transcript_json.md",
+    "think/planner.md",
+    "apps/chat/title.md",
+]
 
 
 # ---------------------------------------------------------------------------
@@ -198,6 +142,49 @@ CONTEXT_DEFAULTS: Dict[str, Dict[str, Any]] = {
 _context_registry: Optional[Dict[str, Dict[str, Any]]] = None
 
 
+def _discover_prompt_contexts() -> Dict[str, Dict[str, Any]]:
+    """Load context metadata from prompt files listed in PROMPT_PATHS.
+
+    Each file must have YAML frontmatter with:
+    - context: The context string (e.g., "observe.enrich")
+    - tier: Tier number (1=pro, 2=flash, 3=lite)
+    - label: Human-readable name
+    - group: Settings UI category
+
+    Returns
+    -------
+    Dict[str, Dict[str, Any]]
+        Mapping of context patterns to {tier, label, group} dicts.
+    """
+    contexts = {}
+    base_dir = Path(__file__).parent.parent  # Project root
+
+    for rel_path in PROMPT_PATHS:
+        path = base_dir / rel_path
+        if not path.exists():
+            logging.getLogger(__name__).warning(f"Prompt file not found: {path}")
+            continue
+
+        try:
+            post = frontmatter.load(path)
+            meta = post.metadata or {}
+
+            context = meta.get("context")
+            if not context:
+                logging.getLogger(__name__).warning(f"No context in {path}")
+                continue
+
+            contexts[context] = {
+                "tier": meta.get("tier", TIER_FLASH),
+                "label": meta.get("label", context),
+                "group": meta.get("group", "Other"),
+            }
+        except Exception as e:
+            logging.getLogger(__name__).warning(f"Failed to load {path}: {e}")
+
+    return contexts
+
+
 def _discover_muse_contexts() -> Dict[str, Dict[str, Any]]:
     """Discover muse context defaults from muse/*.md config files.
 
@@ -265,20 +252,20 @@ def _discover_muse_contexts() -> Dict[str, Dict[str, Any]]:
 
 
 def _build_context_registry() -> Dict[str, Dict[str, Any]]:
-    """Build complete context registry from static defaults and discovered configs.
+    """Build complete context registry from discovered configs.
 
     Merges:
-    1. Static CONTEXT_DEFAULTS (non-discoverable contexts)
+    1. Prompt contexts from _discover_prompt_contexts()
     2. Category contexts from observe/describe.py CATEGORIES
-    3. Agent contexts from _discover_agent_contexts()
+    3. Muse contexts from _discover_muse_contexts()
 
     Returns
     -------
     Dict[str, Dict[str, Any]]
         Complete context registry mapping patterns to {tier, label, group}.
     """
-    # Start with static defaults
-    registry = dict(CONTEXT_DEFAULTS)
+    # Start with prompt contexts (from PROMPT_PATHS)
+    registry = _discover_prompt_contexts()
 
     # Merge category contexts (lazy import to avoid circular dependency)
     try:
@@ -336,7 +323,7 @@ def _resolve_tier(context: str) -> int:
     providers_config = journal_config.get("providers", {})
     contexts = providers_config.get("contexts", {})
 
-    # Get dynamic context registry (includes static defaults + discovered categories/agents)
+    # Get dynamic context registry (discovered prompts, categories, muse configs)
     registry = get_context_registry()
 
     # Check journal config contexts first (exact match)
@@ -496,7 +483,7 @@ def resolve_provider(context: str) -> tuple[str, str]:
 
     # No context match - check dynamic context registry for this context
     if match_config is None:
-        # Get dynamic context registry (includes static defaults + discovered categories/agents)
+        # Get dynamic context registry (discovered prompts, categories, muse configs)
         registry = get_context_registry()
 
         # Check for matching context default (exact match first, then glob)
@@ -1163,7 +1150,7 @@ __all__ = [
     # Provider configuration
     "DEFAULT_TIER",
     "DEFAULT_PROVIDER",
-    "CONTEXT_DEFAULTS",
+    "PROMPT_PATHS",
     "get_context_registry",
     # Model constants (used by provider backends for defaults)
     "GEMINI_FLASH",
diff --git a/think/planner.md b/think/planner.md
index f7b6425b6..761b922d0 100644
--- a/think/planner.md
+++ b/think/planner.md
@@ -1,3 +1,9 @@
+---
+context: planner.generate
+tier: 2
+label: Agent Prompt Generation
+group: Think
+---
 You are a strategic research planner for the solstone journal assistant, specialized in creating comprehensive plans to research and analyze personal journal data to answer user requests.
 
 ## Core Role and Limitations