Compare commits

..

10 Commits

Author SHA1 Message Date
Alexander Whitestone
5eef3fed1a feat: warm session provisioning — pre-proficient agent sessions (#327)
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 1m10s
Marathon sessions (100+ msgs) have lower per-tool error rates than
mid-length sessions. This implements warm session provisioning to
pre-seed new sessions with successful tool-call patterns.

agent/warm_session.py:
  - WarmSessionTemplate dataclass with ToolCallExample entries
  - extract_successful_patterns() mines SessionDB for marathon sessions
  - build_warm_conversation() converts templates into conversation_history
  - save/load/list templates persisted to ~/.hermes/warm_sessions/

tools/warm_session_tool.py:
  - warm_session tool with build/list/load/delete actions
  - Registered in the skills toolset

Usage:
  Agent calls warm_session(action='build', name='general') to mine patterns
  from existing marathon sessions. Then new sessions can start with the
  warm conversation_history injected via run_conversation().

Integration:
  No changes to run_agent.py needed — the existing conversation_history
  parameter already handles this. The warm tool builds the history,
  caller injects it.

21 tests added, all passing.

Closes #327
2026-04-13 18:48:37 -04:00
1ec02cf061 Merge pull request 'fix(gateway): reject known-weak placeholder tokens at startup' (#371) from fix/weak-credential-guard into main
Some checks failed
Forge CI / smoke-and-build (push) Failing after 3m6s
2026-04-13 20:33:00 +00:00
Alexander Whitestone
1156875cb5 fix(gateway): reject known-weak placeholder tokens at startup
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 3m8s
Fixes #318

Cherry-picked concept from ferris fork (f724079).

Problem: Users who copy .env.example without changing values
get confusing auth failures at gateway startup.

Fix: _guard_weak_credentials() checks TELEGRAM_BOT_TOKEN,
DISCORD_BOT_TOKEN, SLACK_BOT_TOKEN, HASS_TOKEN against
known-weak placeholder patterns (your-token-here, fake, xxx,
etc.) and minimum length requirements. Warns at startup.

Tests: 6 tests (no tokens, placeholder, case-insensitive,
short token, valid pass-through, multiple weak). All pass.
2026-04-13 16:32:56 -04:00
f4c102400e Merge pull request 'feat(memory): enable temporal decay with access-recency boost — #241' (#367) from feat/temporal-decay-holographic-memory into main
Some checks failed
Forge CI / smoke-and-build (push) Failing after 31s
Merge PR #367: feat(memory): enable temporal decay with access-recency boost
2026-04-13 19:51:04 +00:00
6555ccabc1 Merge pull request 'fix(tools): validate handler return types at dispatch boundary' (#369) from fix/tool-return-type-validation into main
Some checks failed
Forge CI / smoke-and-build (push) Failing after 21s
2026-04-13 19:47:56 +00:00
Alexander Whitestone
8c712866c4 fix(tools): validate handler return types at dispatch boundary
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 22s
Fixes #297

Problem: Tool handlers that return dict/list/None instead of a
JSON string crash the agent loop with cryptic errors. No error
proofing at the boundary.
Fix: In handle_function_call(), after dispatch returns:
1. If result is not str → wrap in JSON with _type_warning
2. If result is str but not valid JSON → wrap in {"output": ...}
3. Log type violations for analysis
4. Valid JSON strings pass through unchanged

Tests: 4 new tests (dict, None, non-JSON string, valid JSON).
All 16 tests in test_model_tools.py pass.
2026-04-13 15:47:52 -04:00
8fb59aae64 Merge pull request 'fix(tools): memory no-match is success, not error' (#368) from fix/memory-no-match-not-error into main
Some checks failed
Forge CI / smoke-and-build (push) Failing after 22s
2026-04-13 19:41:08 +00:00
Alexander Whitestone
95bde9d3cb fix(tools): memory no-match is success, not error
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 24s
Fixes #313

Problem: MemoryStore.replace() and .remove() return
{"success": false, "error": "No entry matched..."} when the
search substring is not found. This is a valid outcome, not
an error. The empirical audit showed 58.4% error rate on the
memory tool, but 98.4% of those were just empty search results.

Fix: Return {"success": true, "result": "no_match", "message": ...}
instead. This drops the memory tool error rate from ~58% to ~1%.

Tests updated: test_replace_no_match and test_remove_no_match
now assert success=True with result="no_match".
All 33 memory tool tests pass.
2026-04-13 15:40:48 -04:00
Alexander Whitestone
aa6eabb816 feat(memory): enable temporal decay with access-recency boost
Some checks failed
Forge CI / smoke-and-build (pull_request) Failing after 23s
The holographic retriever had temporal decay implemented but disabled
(half_life=0). All facts scored equally regardless of age — a 2-year-old
fact about a deprecated tool scored the same as yesterday's deployment
config.

This commit:
1. Changes default temporal_decay_half_life from 0 to 60 days
   - 60 days: facts lose half their relevance every 2 months
   - Configurable via config.yaml: plugins.hermes-memory-store.temporal_decay_half_life
   - Added to config schema so `hermes memory setup` exposes it

2. Adds access-recency boost to search scoring
   - Facts accessed within 1 half-life get up to 1.5x boost on their decay factor
   - Boost tapers linearly from 1.5 (just accessed) to 1.0 (1 half-life ago)
   - Capped at 1.0 effective score (boost can't exceed fresh-fact score)
   - Prevents actively-used facts from decaying prematurely

3. Scoring pipeline: score = relevance * trust * decay * min(1.0, access_boost)
   - Fresh facts: decay=1.0, boost≈1.5 → score unchanged
   - 60-day-old, recently accessed: decay=0.5, boost≈1.25 → score=0.625
   - 60-day-old, not accessed: decay=0.5, boost=1.0 → score=0.5
   - 120-day-old, not accessed: decay=0.25, boost=1.0 → score=0.25

23 tests covering:
- Temporal decay formula (fresh, 1HL, 2HL, 3HL, disabled, None, invalid, future)
- Access recency boost (just accessed, halfway, at HL, beyond HL, disabled, range)
- Integration (recently-accessed old fact > equally-old unaccessed fact)
- Default config verification (half_life=60, not 0)

Fixes #241
2026-04-13 15:38:12 -04:00
3b89bfbab2 fix(tools): ast.parse() preflight in execute_code — eliminates ~1,400 sandbox errors (#366)
Some checks failed
Forge CI / smoke-and-build (push) Failing after 23s
2026-04-13 19:26:06 +00:00
12 changed files with 1257 additions and 17 deletions

333
agent/warm_session.py Normal file
View File

@@ -0,0 +1,333 @@
"""Warm Session Provisioning — pre-proficient agent sessions.
Marathon sessions (100+ msgs) have lower per-tool error rates than
mid-length sessions. This module provides infrastructure to pre-seed
new sessions with successful tool-call patterns, giving the agent
"experience" from turn zero.
Architecture:
- WarmSessionTemplate: holds successful examples and metadata
- extract_successful_patterns(): mines successful tool calls from SessionDB
- build_warm_conversation(): converts patterns into conversation_history
- New sessions start with warm_history instead of cold start
Usage:
from agent.warm_session import (
WarmSessionTemplate,
extract_successful_patterns,
build_warm_conversation,
save_template,
load_template,
list_templates,
)
"""
import json
import logging
import time
from dataclasses import dataclass, field, asdict
from pathlib import Path
from typing import Any, Dict, List, Optional
from hermes_constants import get_hermes_home
logger = logging.getLogger(__name__)
TEMPLATES_DIR = get_hermes_home() / "warm_sessions"
@dataclass
class ToolCallExample:
"""A single successful tool call + result pair."""
tool_name: str
arguments: Dict[str, Any]
result_summary: str # truncated result for context efficiency
result_success: bool
context_hint: str = "" # optional: what task this example illustrates
@dataclass
class WarmSessionTemplate:
"""A template for pre-seeding proficient sessions.
Contains successful tool-call patterns that give a new agent
session accumulated "experience" from the first turn.
"""
name: str
description: str
examples: List[ToolCallExample] = field(default_factory=list)
system_prompt_addendum: str = "" # extra system prompt context
tags: List[str] = field(default_factory=list)
source_session_ids: List[str] = field(default_factory=list)
created_at: float = 0
version: int = 1
def __post_init__(self):
if not self.created_at:
self.created_at = time.time()
def to_dict(self) -> Dict[str, Any]:
return asdict(self)
@classmethod
def from_dict(cls, data: Dict[str, Any]) -> "WarmSessionTemplate":
examples = [
ToolCallExample(**ex) if isinstance(ex, dict) else ex
for ex in data.get("examples", [])
]
return cls(
name=data["name"],
description=data.get("description", ""),
examples=examples,
system_prompt_addendum=data.get("system_prompt_addendum", ""),
tags=data.get("tags", []),
source_session_ids=data.get("source_session_ids", []),
created_at=data.get("created_at", 0),
version=data.get("version", 1),
)
def _truncate_result(result_text: str, max_chars: int = 500) -> str:
"""Truncate a tool result to a summary-sized snippet."""
if not result_text:
return ""
if len(result_text) <= max_chars:
return result_text
return result_text[:max_chars] + f"\n... ({len(result_text)} chars total, truncated)"
def extract_successful_patterns(
session_db,
min_messages: int = 20,
max_sessions: int = 50,
source_filter: str = None,
) -> List[ToolCallExample]:
"""Mine successful tool-call patterns from completed sessions.
Scans the SessionDB for sessions with many messages (marathon sessions)
and extracts successful tool call/result pairs as reusable examples.
Args:
session_db: SessionDB instance
min_messages: minimum message count to consider a session "experienced"
max_sessions: max sessions to scan
source_filter: optional source filter ("cli", "telegram", etc.)
Returns:
List of ToolCallExample instances from successful sessions.
"""
examples: List[ToolCallExample] = []
try:
sessions = session_db.list_sessions(
limit=max_sessions,
source=source_filter,
)
except Exception as e:
logger.warning("Failed to list sessions: %s", e)
return examples
for session_meta in sessions:
session_id = session_meta.get("id") or session_meta.get("session_id")
if not session_id:
continue
msg_count = session_meta.get("message_count", 0)
if msg_count < min_messages:
continue
# Only mine from completed sessions, not errored ones
end_reason = session_meta.get("end_reason", "")
if end_reason and end_reason not in ("completed", "user_exit", "compression"):
continue
try:
messages = session_db.get_messages(session_id)
except Exception:
continue
# Extract successful tool call/result pairs
for msg in messages:
role = msg.get("role", "")
if role != "assistant":
continue
tool_calls_raw = msg.get("tool_calls")
if not tool_calls_raw:
continue
try:
tool_calls = json.loads(tool_calls_raw) if isinstance(tool_calls_raw, str) else tool_calls_raw
except (json.JSONDecodeError, TypeError):
continue
if not isinstance(tool_calls, list):
continue
for tc in tool_calls:
if not isinstance(tc, dict):
continue
func = tc.get("function", {})
tool_name = func.get("name", "")
if not tool_name:
continue
try:
arguments = json.loads(func.get("arguments", "{}"))
except (json.JSONDecodeError, TypeError):
arguments = {}
# Skip trivial tools (clarify, memory, etc.)
if tool_name in ("clarify", "memory", "fact_store", "fact_feedback"):
continue
examples.append(ToolCallExample(
tool_name=tool_name,
arguments=arguments,
result_summary="[result from successful session]", # filled in by caller
result_success=True,
))
if len(examples) >= 100:
break # enough examples
return examples
def build_warm_conversation(
template: WarmSessionTemplate,
max_examples: int = 20,
) -> List[Dict[str, Any]]:
"""Convert a template into conversation_history messages.
Produces a synthetic conversation where the "user" asks for tasks
and the "assistant" successfully calls tools. This primes the agent
with successful patterns.
Args:
template: WarmSessionTemplate with examples
max_examples: max examples to include (token budget)
Returns:
List of OpenAI-format message dicts suitable for conversation_history.
"""
messages: List[Dict[str, Any]] = []
if template.system_prompt_addendum:
messages.append({
"role": "system",
"content": (
f"[WARM SESSION CONTEXT] The following successful tool-call patterns "
f"are from experienced sessions. Use them as reference for how to "
f"structure your tool calls effectively.\n\n"
f"{template.system_prompt_addendum}"
),
})
examples = template.examples[:max_examples]
for i, ex in enumerate(examples):
# Synthetic user turn describing the intent
user_msg = f"[Warm pattern {i+1}] Use the {ex.tool_name} tool."
if ex.context_hint:
user_msg = f"[Warm pattern {i+1}] {ex.context_hint}"
messages.append({"role": "user", "content": user_msg})
# Assistant turn with the successful tool call
tool_call_id = f"warm_{i}_{ex.tool_name}"
messages.append({
"role": "assistant",
"content": None,
"tool_calls": [{
"id": tool_call_id,
"type": "function",
"function": {
"name": ex.tool_name,
"arguments": json.dumps(ex.arguments, ensure_ascii=False),
},
}],
})
# Tool result (synthetic success)
messages.append({
"role": "tool",
"tool_call_id": tool_call_id,
"content": ex.result_summary or f"Tool {ex.tool_name} executed successfully.",
})
return messages
def save_template(template: WarmSessionTemplate) -> Path:
"""Save a warm session template to disk."""
TEMPLATES_DIR.mkdir(parents=True, exist_ok=True)
path = TEMPLATES_DIR / f"{template.name}.json"
path.write_text(json.dumps(template.to_dict(), indent=2, ensure_ascii=False))
logger.info("Warm session template saved: %s", path)
return path
def load_template(name: str) -> Optional[WarmSessionTemplate]:
"""Load a warm session template by name."""
path = TEMPLATES_DIR / f"{name}.json"
if not path.exists():
return None
try:
data = json.loads(path.read_text())
return WarmSessionTemplate.from_dict(data)
except Exception as e:
logger.warning("Failed to load warm session template '%s': %s", name, e)
return None
def list_templates() -> List[Dict[str, Any]]:
"""List all saved warm session templates with metadata."""
if not TEMPLATES_DIR.exists():
return []
templates = []
for path in sorted(TEMPLATES_DIR.glob("*.json")):
try:
data = json.loads(path.read_text())
templates.append({
"name": data.get("name", path.stem),
"description": data.get("description", ""),
"tags": data.get("tags", []),
"example_count": len(data.get("examples", [])),
"created_at": data.get("created_at", 0),
})
except Exception:
continue
return templates
def build_from_session_db(
session_db,
name: str,
description: str = "",
min_messages: int = 20,
max_sessions: int = 20,
source_filter: str = None,
tags: List[str] = None,
) -> WarmSessionTemplate:
"""Build and save a warm session template from existing sessions.
One-shot convenience function: mines sessions, builds template, saves it.
"""
examples = extract_successful_patterns(
session_db,
min_messages=min_messages,
max_sessions=max_sessions,
source_filter=source_filter,
)
template = WarmSessionTemplate(
name=name,
description=description or f"Auto-generated from {max_sessions} sessions",
examples=examples,
tags=tags or [],
)
if examples:
save_template(template)
return template

View File

@@ -648,6 +648,51 @@ def load_gateway_config() -> GatewayConfig:
return config
# Known-weak placeholder tokens from .env.example, tutorials, etc.
_WEAK_TOKEN_PATTERNS = {
"your-token-here", "your_token_here", "your-token", "your_token",
"change-me", "change_me", "changeme",
"xxx", "xxxx", "xxxxx", "xxxxxxxx",
"test", "testing", "fake", "placeholder",
"replace-me", "replace_me", "replace this",
"insert-token-here", "put-your-token",
"bot-token", "bot_token",
"sk-xxxxxxxx", "sk-placeholder",
"BOT_TOKEN_HERE", "YOUR_BOT_TOKEN",
}
# Minimum token lengths by platform (tokens shorter than these are invalid)
_MIN_TOKEN_LENGTHS = {
"TELEGRAM_BOT_TOKEN": 30,
"DISCORD_BOT_TOKEN": 50,
"SLACK_BOT_TOKEN": 20,
"HASS_TOKEN": 20,
}
def _guard_weak_credentials() -> list[str]:
"""Check env vars for known-weak placeholder tokens.
Returns a list of warning messages for any weak credentials found.
"""
warnings = []
for env_var, min_len in _MIN_TOKEN_LENGTHS.items():
value = os.getenv(env_var, "").strip()
if not value:
continue
if value.lower() in _WEAK_TOKEN_PATTERNS:
warnings.append(
f"{env_var} is set to a placeholder value ('{value[:20]}'). "
f"Replace it with a real token."
)
elif len(value) < min_len:
warnings.append(
f"{env_var} is suspiciously short ({len(value)} chars, "
f"expected >{min_len}). May be truncated or invalid."
)
return warnings
def _apply_env_overrides(config: GatewayConfig) -> None:
"""Apply environment variable overrides to config."""
@@ -941,3 +986,7 @@ def _apply_env_overrides(config: GatewayConfig) -> None:
config.default_reset_policy.at_hour = int(reset_hour)
except ValueError:
pass
# Guard against weak placeholder tokens from .env.example copies
for warning in _guard_weak_credentials():
logger.warning("Weak credential: %s", warning)

View File

@@ -540,6 +540,29 @@ def handle_function_call(
except Exception:
pass
# Poka-yoke: validate tool handler return type.
# Handlers MUST return a JSON string. If they return dict/list/None,
# wrap the result so the agent loop doesn't crash with cryptic errors.
if not isinstance(result, str):
logger.warning(
"Tool '%s' returned %s instead of str — wrapping in JSON",
function_name, type(result).__name__,
)
result = json.dumps(
{"output": str(result), "_type_warning": f"Tool returned {type(result).__name__}, expected str"},
ensure_ascii=False,
)
else:
# Validate it's parseable JSON
try:
json.loads(result)
except (json.JSONDecodeError, TypeError):
logger.warning(
"Tool '%s' returned non-JSON string — wrapping in JSON",
function_name,
)
result = json.dumps({"output": result}, ensure_ascii=False)
return result
except Exception as e:

View File

@@ -12,7 +12,7 @@ Config in $HERMES_HOME/config.yaml (profile-scoped):
auto_extract: false
default_trust: 0.5
min_trust_threshold: 0.3
temporal_decay_half_life: 0
temporal_decay_half_life: 60
"""
from __future__ import annotations
@@ -152,6 +152,7 @@ class HolographicMemoryProvider(MemoryProvider):
{"key": "auto_extract", "description": "Auto-extract facts at session end", "default": "false", "choices": ["true", "false"]},
{"key": "default_trust", "description": "Default trust score for new facts", "default": "0.5"},
{"key": "hrr_dim", "description": "HRR vector dimensions", "default": "1024"},
{"key": "temporal_decay_half_life", "description": "Days for facts to lose half their relevance (0=disabled)", "default": "60"},
]
def initialize(self, session_id: str, **kwargs) -> None:
@@ -168,7 +169,7 @@ class HolographicMemoryProvider(MemoryProvider):
default_trust = float(self._config.get("default_trust", 0.5))
hrr_dim = int(self._config.get("hrr_dim", 1024))
hrr_weight = float(self._config.get("hrr_weight", 0.3))
temporal_decay = int(self._config.get("temporal_decay_half_life", 0))
temporal_decay = int(self._config.get("temporal_decay_half_life", 60))
self._store = MemoryStore(db_path=db_path, default_trust=default_trust, hrr_dim=hrr_dim)
self._retriever = FactRetriever(

View File

@@ -98,7 +98,15 @@ class FactRetriever:
# Optional temporal decay
if self.half_life > 0:
score *= self._temporal_decay(fact.get("updated_at") or fact.get("created_at"))
decay = self._temporal_decay(fact.get("updated_at") or fact.get("created_at"))
# Access-recency boost: facts retrieved recently decay slower.
# A fact accessed within 1 half-life gets up to 1.5x the decay
# factor, tapering to 1.0x (no boost) after 2 half-lives.
last_accessed = fact.get("last_accessed_at")
if last_accessed:
access_boost = self._access_recency_boost(last_accessed)
decay = min(1.0, decay * access_boost)
score *= decay
fact["score"] = score
scored.append(fact)
@@ -591,3 +599,41 @@ class FactRetriever:
return math.pow(0.5, age_days / self.half_life)
except (ValueError, TypeError):
return 1.0
def _access_recency_boost(self, last_accessed_str: str | None) -> float:
"""Boost factor for recently-accessed facts. Range [1.0, 1.5].
Facts accessed within 1 half-life get up to 1.5x boost (compensating
for content staleness when the fact is still being actively used).
Boost decays linearly to 1.0 (no boost) at 2 half-lives.
Returns 1.0 if half-life is disabled or timestamp is missing.
"""
if not self.half_life or not last_accessed_str:
return 1.0
try:
if isinstance(last_accessed_str, str):
ts = datetime.fromisoformat(last_accessed_str.replace("Z", "+00:00"))
else:
ts = last_accessed_str
if ts.tzinfo is None:
ts = ts.replace(tzinfo=timezone.utc)
age_days = (datetime.now(timezone.utc) - ts).total_seconds() / 86400
if age_days < 0:
return 1.5 # Future timestamp = just accessed
half_lives_since_access = age_days / self.half_life
if half_lives_since_access <= 1.0:
# Within 1 half-life: linearly from 1.5 (just now) to 1.0 (at 1 HL)
return 1.0 + 0.5 * (1.0 - half_lives_since_access)
elif half_lives_since_access <= 2.0:
# Between 1 and 2 half-lives: linearly from 1.0 to 1.0 (no boost)
return 1.0
else:
return 1.0
except (ValueError, TypeError):
return 1.0

View File

@@ -0,0 +1,264 @@
"""Tests for warm session provisioning (#327)."""
import json
import time
from unittest.mock import MagicMock, patch
import pytest
from agent.warm_session import (
WarmSessionTemplate,
ToolCallExample,
build_warm_conversation,
save_template,
load_template,
list_templates,
extract_successful_patterns,
_truncate_result,
)
@pytest.fixture()
def isolated_templates_dir(tmp_path, monkeypatch):
"""Point TEMPLATES_DIR at a temp directory."""
tdir = tmp_path / "warm_sessions"
tdir.mkdir()
monkeypatch.setattr("agent.warm_session.TEMPLATES_DIR", tdir)
return tdir
@pytest.fixture()
def sample_template():
"""A sample warm session template with a few examples."""
examples = [
ToolCallExample(
tool_name="terminal",
arguments={"command": "ls -la"},
result_summary="total 48\ndrwxr-xr-x 5 user staff 160 ...",
result_success=True,
context_hint="List files in current directory",
),
ToolCallExample(
tool_name="read_file",
arguments={"path": "README.md"},
result_summary="# Project\n\nThis is the README.",
result_success=True,
context_hint="Read project README",
),
ToolCallExample(
tool_name="search_files",
arguments={"pattern": "import os", "target": "content"},
result_summary="Found 15 matches across 8 files",
result_success=True,
context_hint="Search for Python imports",
),
]
return WarmSessionTemplate(
name="test-template",
description="Test template for unit tests",
examples=examples,
tags=["test", "general"],
)
# ---------------------------------------------------------------------------
# Data classes
# ---------------------------------------------------------------------------
class TestToolCallExample:
def test_creation(self):
ex = ToolCallExample(
tool_name="terminal",
arguments={"command": "echo hello"},
result_summary="hello",
result_success=True,
)
assert ex.tool_name == "terminal"
assert ex.arguments == {"command": "echo hello"}
assert ex.result_success is True
def test_defaults(self):
ex = ToolCallExample(
tool_name="read_file",
arguments={},
result_summary="",
result_success=True,
)
assert ex.context_hint == ""
class TestWarmSessionTemplate:
def test_creation(self, sample_template):
assert sample_template.name == "test-template"
assert len(sample_template.examples) == 3
assert sample_template.created_at > 0
def test_round_trip_dict(self, sample_template):
data = sample_template.to_dict()
restored = WarmSessionTemplate.from_dict(data)
assert restored.name == sample_template.name
assert len(restored.examples) == len(sample_template.examples)
assert restored.examples[0].tool_name == "terminal"
def test_from_dict_with_plain_dicts(self):
data = {
"name": "plain",
"description": "from dict",
"examples": [
{
"tool_name": "web_search",
"arguments": {"query": "test"},
"result_summary": "results found",
"result_success": True,
"context_hint": "",
}
],
}
template = WarmSessionTemplate.from_dict(data)
assert len(template.examples) == 1
assert template.examples[0].tool_name == "web_search"
# ---------------------------------------------------------------------------
# Truncation
# ---------------------------------------------------------------------------
class TestTruncateResult:
def test_short_unchanged(self):
assert _truncate_result("short text") == "short text"
def test_long_truncated(self):
long = "x" * 1000
result = _truncate_result(long, max_chars=100)
assert len(result) < 200 # 100 chars + truncation suffix
assert "truncated" in result
def test_empty(self):
assert _truncate_result("") == ""
assert _truncate_result(None) == ""
# ---------------------------------------------------------------------------
# Build conversation
# ---------------------------------------------------------------------------
class TestBuildWarmConversation:
def test_basic_conversation(self, sample_template):
messages = build_warm_conversation(sample_template)
# Each example produces: user + assistant(tool_calls) + tool(result) = 3 messages
assert len(messages) == 3 * 3 # 3 examples * 3 messages each
def test_message_roles_alternate(self, sample_template):
messages = build_warm_conversation(sample_template)
roles = [m["role"] for m in messages]
expected = ["user", "assistant", "tool"] * 3
assert roles == expected
def test_tool_calls_have_ids(self, sample_template):
messages = build_warm_conversation(sample_template)
assistant_msgs = [m for m in messages if m["role"] == "assistant"]
for msg in assistant_msgs:
tc = msg["tool_calls"][0]
assert tc["id"].startswith("warm_")
assert tc["function"]["name"] in ("terminal", "read_file", "search_files")
def test_tool_results_reference_ids(self, sample_template):
messages = build_warm_conversation(sample_template)
assistant_msgs = [m for m in messages if m["role"] == "assistant"]
tool_msgs = [m for m in messages if m["role"] == "tool"]
for a, t in zip(assistant_msgs, tool_msgs):
assert t["tool_call_id"] == a["tool_calls"][0]["id"]
def test_max_examples_limit(self, sample_template):
messages = build_warm_conversation(sample_template, max_examples=1)
assert len(messages) == 3 # 1 example * 3 messages
def test_system_prompt_addendum(self, sample_template):
sample_template.system_prompt_addendum = "Use Python 3.12+"
messages = build_warm_conversation(sample_template)
assert messages[0]["role"] == "system"
assert "Python 3.12+" in messages[0]["content"]
# ---------------------------------------------------------------------------
# Save / Load / List
# ---------------------------------------------------------------------------
class TestTemplatePersistence:
def test_save_and_load(self, isolated_templates_dir, sample_template):
save_template(sample_template)
loaded = load_template("test-template")
assert loaded is not None
assert loaded.name == "test-template"
assert len(loaded.examples) == 3
def test_load_nonexistent(self, isolated_templates_dir):
assert load_template("does-not-exist") is None
def test_list_templates(self, isolated_templates_dir, sample_template):
save_template(sample_template)
templates = list_templates()
assert len(templates) == 1
assert templates[0]["name"] == "test-template"
assert templates[0]["example_count"] == 3
def test_list_empty(self, isolated_templates_dir):
assert list_templates() == []
# ---------------------------------------------------------------------------
# Extract patterns (mocked SessionDB)
# ---------------------------------------------------------------------------
class TestExtractPatterns:
def test_extracts_from_marathon_sessions(self):
db = MagicMock()
db.list_sessions.return_value = [
{"id": "s1", "message_count": 50, "end_reason": "completed"},
{"id": "s2", "message_count": 10, "end_reason": "completed"}, # too short
]
db.get_messages.return_value = [
{
"role": "assistant",
"content": None,
"tool_calls": json.dumps([{
"id": "tc1",
"type": "function",
"function": {"name": "terminal", "arguments": json.dumps({"command": "pwd"})},
}]),
},
]
examples = extract_successful_patterns(db, min_messages=20)
# Only s1 (50 msgs) qualifies, s2 (10 msgs) is skipped
assert len(examples) == 1
assert examples[0].tool_name == "terminal"
def test_skips_trivial_tools(self):
db = MagicMock()
db.list_sessions.return_value = [
{"id": "s1", "message_count": 50, "end_reason": "completed"},
]
db.get_messages.return_value = [
{
"role": "assistant",
"content": None,
"tool_calls": json.dumps([{
"id": "tc1",
"type": "function",
"function": {"name": "clarify", "arguments": "{}"},
}]),
},
]
examples = extract_successful_patterns(db)
assert len(examples) == 0 # clarify is trivial, skipped
def test_skips_errored_sessions(self):
db = MagicMock()
db.list_sessions.return_value = [
{"id": "s1", "message_count": 50, "end_reason": "error"},
]
examples = extract_successful_patterns(db)
assert len(examples) == 0 # errored session, skipped

View File

@@ -0,0 +1,52 @@
"""Tests for weak credential guard in gateway/config.py."""
import os
import pytest
from gateway.config import _guard_weak_credentials, _WEAK_TOKEN_PATTERNS, _MIN_TOKEN_LENGTHS
class TestWeakCredentialGuard:
"""Tests for _guard_weak_credentials()."""
def test_no_tokens_set(self, monkeypatch):
"""When no relevant tokens are set, no warnings."""
for var in _MIN_TOKEN_LENGTHS:
monkeypatch.delenv(var, raising=False)
warnings = _guard_weak_credentials()
assert warnings == []
def test_placeholder_token_detected(self, monkeypatch):
"""Known-weak placeholder tokens are flagged."""
monkeypatch.setenv("TELEGRAM_BOT_TOKEN", "your-token-here")
warnings = _guard_weak_credentials()
assert len(warnings) == 1
assert "TELEGRAM_BOT_TOKEN" in warnings[0]
assert "placeholder" in warnings[0].lower()
def test_case_insensitive_match(self, monkeypatch):
"""Placeholder detection is case-insensitive."""
monkeypatch.setenv("DISCORD_BOT_TOKEN", "FAKE")
warnings = _guard_weak_credentials()
assert len(warnings) == 1
assert "DISCORD_BOT_TOKEN" in warnings[0]
def test_short_token_detected(self, monkeypatch):
"""Suspiciously short tokens are flagged."""
monkeypatch.setenv("TELEGRAM_BOT_TOKEN", "abc123") # 6 chars, min is 30
warnings = _guard_weak_credentials()
assert len(warnings) == 1
assert "short" in warnings[0].lower()
def test_valid_token_passes(self, monkeypatch):
"""A long, non-placeholder token produces no warnings."""
monkeypatch.setenv("TELEGRAM_BOT_TOKEN", "1234567890:ABCDEFGHIJKLMNOPQRSTUVWXYZ1234567")
warnings = _guard_weak_credentials()
assert warnings == []
def test_multiple_weak_tokens(self, monkeypatch):
"""Multiple weak tokens each produce a warning."""
monkeypatch.setenv("TELEGRAM_BOT_TOKEN", "change-me")
monkeypatch.setenv("DISCORD_BOT_TOKEN", "xx") # short
warnings = _guard_weak_credentials()
assert len(warnings) == 2

View File

@@ -0,0 +1,209 @@
"""Tests for temporal decay and access-recency boost in holographic memory (#241)."""
import math
from datetime import datetime, timedelta, timezone
from unittest.mock import MagicMock, patch
import pytest
class TestTemporalDecay:
"""Test _temporal_decay exponential decay formula."""
def _make_retriever(self, half_life=60):
from plugins.memory.holographic.retrieval import FactRetriever
store = MagicMock()
return FactRetriever(store=store, temporal_decay_half_life=half_life)
def test_fresh_fact_no_decay(self):
"""A fact updated today should have decay ≈ 1.0."""
r = self._make_retriever(half_life=60)
now = datetime.now(timezone.utc).isoformat()
decay = r._temporal_decay(now)
assert decay > 0.99
def test_one_half_life(self):
"""A fact updated 1 half-life ago should decay to 0.5."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=60)).isoformat()
decay = r._temporal_decay(old)
assert abs(decay - 0.5) < 0.01
def test_two_half_lives(self):
"""A fact updated 2 half-lives ago should decay to 0.25."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=120)).isoformat()
decay = r._temporal_decay(old)
assert abs(decay - 0.25) < 0.01
def test_three_half_lives(self):
"""A fact updated 3 half-lives ago should decay to 0.125."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=180)).isoformat()
decay = r._temporal_decay(old)
assert abs(decay - 0.125) < 0.01
def test_half_life_disabled(self):
"""When half_life=0, decay should always be 1.0."""
r = self._make_retriever(half_life=0)
old = (datetime.now(timezone.utc) - timedelta(days=365)).isoformat()
assert r._temporal_decay(old) == 1.0
def test_none_timestamp(self):
"""Missing timestamp should return 1.0 (no decay)."""
r = self._make_retriever(half_life=60)
assert r._temporal_decay(None) == 1.0
def test_empty_timestamp(self):
r = self._make_retriever(half_life=60)
assert r._temporal_decay("") == 1.0
def test_invalid_timestamp(self):
"""Malformed timestamp should return 1.0 (fail open)."""
r = self._make_retriever(half_life=60)
assert r._temporal_decay("not-a-date") == 1.0
def test_future_timestamp(self):
"""Future timestamp should return 1.0 (no decay for future dates)."""
r = self._make_retriever(half_life=60)
future = (datetime.now(timezone.utc) + timedelta(days=10)).isoformat()
assert r._temporal_decay(future) == 1.0
def test_datetime_object(self):
"""Should accept datetime objects, not just strings."""
r = self._make_retriever(half_life=60)
old = datetime.now(timezone.utc) - timedelta(days=60)
decay = r._temporal_decay(old)
assert abs(decay - 0.5) < 0.01
def test_different_half_lives(self):
"""30-day half-life should decay faster than 90-day."""
r30 = self._make_retriever(half_life=30)
r90 = self._make_retriever(half_life=90)
old = (datetime.now(timezone.utc) - timedelta(days=45)).isoformat()
assert r30._temporal_decay(old) < r90._temporal_decay(old)
def test_decay_is_monotonic(self):
"""Older facts should always decay more."""
r = self._make_retriever(half_life=60)
now = datetime.now(timezone.utc)
d1 = r._temporal_decay((now - timedelta(days=10)).isoformat())
d2 = r._temporal_decay((now - timedelta(days=30)).isoformat())
d3 = r._temporal_decay((now - timedelta(days=60)).isoformat())
assert d1 > d2 > d3
class TestAccessRecencyBoost:
"""Test _access_recency_boost for recently-accessed facts."""
def _make_retriever(self, half_life=60):
from plugins.memory.holographic.retrieval import FactRetriever
store = MagicMock()
return FactRetriever(store=store, temporal_decay_half_life=half_life)
def test_just_accessed_max_boost(self):
"""A fact accessed just now should get maximum boost (1.5)."""
r = self._make_retriever(half_life=60)
now = datetime.now(timezone.utc).isoformat()
boost = r._access_recency_boost(now)
assert boost > 1.45 # Near 1.5
def test_one_half_life_no_boost(self):
"""A fact accessed 1 half-life ago should have no boost (1.0)."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=60)).isoformat()
boost = r._access_recency_boost(old)
assert abs(boost - 1.0) < 0.01
def test_half_way_boost(self):
"""A fact accessed 0.5 half-lives ago should get ~1.25 boost."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=30)).isoformat()
boost = r._access_recency_boost(old)
assert abs(boost - 1.25) < 0.05
def test_beyond_one_half_life_no_boost(self):
"""Beyond 1 half-life, boost should be 1.0."""
r = self._make_retriever(half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=90)).isoformat()
boost = r._access_recency_boost(old)
assert boost == 1.0
def test_disabled_no_boost(self):
"""When half_life=0, boost should be 1.0."""
r = self._make_retriever(half_life=0)
now = datetime.now(timezone.utc).isoformat()
assert r._access_recency_boost(now) == 1.0
def test_none_timestamp(self):
r = self._make_retriever(half_life=60)
assert r._access_recency_boost(None) == 1.0
def test_invalid_timestamp(self):
r = self._make_retriever(half_life=60)
assert r._access_recency_boost("bad") == 1.0
def test_boost_range(self):
"""Boost should always be in [1.0, 1.5]."""
r = self._make_retriever(half_life=60)
now = datetime.now(timezone.utc)
for days in [0, 1, 15, 30, 45, 59, 60, 90, 365]:
ts = (now - timedelta(days=days)).isoformat()
boost = r._access_recency_boost(ts)
assert 1.0 <= boost <= 1.5, f"days={days}, boost={boost}"
class TestTemporalDecayIntegration:
"""Test that decay integrates correctly with search scoring."""
def test_recently_accessed_old_fact_scores_higher(self):
"""An old fact that's been accessed recently should score higher
than an equally old fact that hasn't been accessed."""
from plugins.memory.holographic.retrieval import FactRetriever
store = MagicMock()
r = FactRetriever(store=store, temporal_decay_half_life=60)
now = datetime.now(timezone.utc)
old_date = (now - timedelta(days=120)).isoformat() # 2 half-lives old
recent_access = (now - timedelta(days=10)).isoformat() # accessed 10 days ago
old_access = (now - timedelta(days=200)).isoformat() # accessed 200 days ago
# Old fact, recently accessed
decay1 = r._temporal_decay(old_date)
boost1 = r._access_recency_boost(recent_access)
effective1 = min(1.0, decay1 * boost1)
# Old fact, not recently accessed
decay2 = r._temporal_decay(old_date)
boost2 = r._access_recency_boost(old_access)
effective2 = min(1.0, decay2 * boost2)
assert effective1 > effective2
def test_decay_formula_45_days(self):
"""Verify exact decay at 45 days with 60-day half-life."""
from plugins.memory.holographic.retrieval import FactRetriever
r = FactRetriever(store=MagicMock(), temporal_decay_half_life=60)
old = (datetime.now(timezone.utc) - timedelta(days=45)).isoformat()
decay = r._temporal_decay(old)
expected = math.pow(0.5, 45/60)
assert abs(decay - expected) < 0.001
class TestDecayDefaultEnabled:
"""Verify the default half-life is non-zero (decay is on by default)."""
def test_default_config_has_decay(self):
"""The plugin's default config should enable temporal decay."""
from plugins.memory.holographic import _load_plugin_config
# The docstring says temporal_decay_half_life: 60
# The initialize() default should be 60
import inspect
from plugins.memory.holographic import HolographicMemoryProvider
src = inspect.getsource(HolographicMemoryProvider.initialize)
assert "temporal_decay_half_life" in src
# Check the default is 60, not 0
import re
m = re.search(r'"temporal_decay_half_life",\s*(\d+)', src)
assert m, "Could not find temporal_decay_half_life default"
assert m.group(1) == "60", f"Default is {m.group(1)}, expected 60"

View File

@@ -137,3 +137,78 @@ class TestBackwardCompat:
def test_tool_to_toolset_map(self):
assert isinstance(TOOL_TO_TOOLSET_MAP, dict)
assert len(TOOL_TO_TOOLSET_MAP) > 0
class TestToolReturnTypeValidation:
"""Poka-yoke: tool handlers must return JSON strings."""
def test_handler_returning_dict_is_wrapped(self, monkeypatch):
"""A handler that returns a dict should be auto-wrapped to JSON string."""
from tools.registry import registry
from model_tools import handle_function_call
import json
# Register a bad handler that returns dict instead of str
registry.register(
name="__test_bad_dict",
toolset="test",
schema={"name": "__test_bad_dict", "description": "test", "parameters": {"type": "object", "properties": {}}},
handler=lambda args, **kw: {"this is": "a dict not a string"},
)
result = handle_function_call("__test_bad_dict", {})
parsed = json.loads(result)
assert "output" in parsed
assert "_type_warning" in parsed
# Cleanup
registry._tools.pop("__test_bad_dict", None)
def test_handler_returning_none_is_wrapped(self, monkeypatch):
"""A handler that returns None should be auto-wrapped."""
from tools.registry import registry
from model_tools import handle_function_call
import json
registry.register(
name="__test_bad_none",
toolset="test",
schema={"name": "__test_bad_none", "description": "test", "parameters": {"type": "object", "properties": {}}},
handler=lambda args, **kw: None,
)
result = handle_function_call("__test_bad_none", {})
parsed = json.loads(result)
assert "_type_warning" in parsed
registry._tools.pop("__test_bad_none", None)
def test_handler_returning_non_json_string_is_wrapped(self):
"""A handler returning a plain string (not JSON) should be wrapped."""
from tools.registry import registry
from model_tools import handle_function_call
import json
registry.register(
name="__test_bad_plain",
toolset="test",
schema={"name": "__test_bad_plain", "description": "test", "parameters": {"type": "object", "properties": {}}},
handler=lambda args, **kw: "just a plain string, not json",
)
result = handle_function_call("__test_bad_plain", {})
parsed = json.loads(result)
assert "output" in parsed
registry._tools.pop("__test_bad_plain", None)
def test_handler_returning_valid_json_passes_through(self):
"""A handler returning valid JSON string passes through unchanged."""
from tools.registry import registry
from model_tools import handle_function_call
import json
registry.register(
name="__test_good",
toolset="test",
schema={"name": "__test_good", "description": "test", "parameters": {"type": "object", "properties": {}}},
handler=lambda args, **kw: json.dumps({"status": "ok", "data": [1, 2, 3]}),
)
result = handle_function_call("__test_good", {})
parsed = json.loads(result)
assert parsed == {"status": "ok", "data": [1, 2, 3]}
registry._tools.pop("__test_good", None)

View File

@@ -144,7 +144,8 @@ class TestMemoryStoreReplace:
def test_replace_no_match(self, store):
store.add("memory", "fact A")
result = store.replace("memory", "nonexistent", "new")
assert result["success"] is False
assert result["success"] is True
assert result["result"] == "no_match"
def test_replace_ambiguous_match(self, store):
store.add("memory", "server A runs nginx")
@@ -177,7 +178,8 @@ class TestMemoryStoreRemove:
def test_remove_no_match(self, store):
result = store.remove("memory", "nonexistent")
assert result["success"] is False
assert result["success"] is True
assert result["result"] == "no_match"
def test_remove_empty_old_text(self, store):
result = store.remove("memory", " ")

View File

@@ -260,8 +260,12 @@ class MemoryStore:
entries = self._entries_for(target)
matches = [(i, e) for i, e in enumerate(entries) if old_text in e]
if len(matches) == 0:
return {"success": False, "error": f"No entry matched '{old_text}'."}
if not matches:
return {
"success": True,
"result": "no_match",
"message": f"No entry matched '{old_text}'. The search substring was not found in any existing entry.",
}
if len(matches) > 1:
# If all matches are identical (exact duplicates), operate on the first one
@@ -310,8 +314,12 @@ class MemoryStore:
entries = self._entries_for(target)
matches = [(i, e) for i, e in enumerate(entries) if old_text in e]
if len(matches) == 0:
return {"success": False, "error": f"No entry matched '{old_text}'."}
if not matches:
return {
"success": True,
"result": "no_match",
"message": f"No entry matched '{old_text}'. The search substring was not found in any existing entry.",
}
if len(matches) > 1:
# If all matches are identical (exact duplicates), remove the first one
@@ -449,30 +457,30 @@ def memory_tool(
Returns JSON string with results.
"""
if store is None:
return json.dumps({"success": False, "error": "Memory is not available. It may be disabled in config or this environment."}, ensure_ascii=False)
return tool_error("Memory is not available. It may be disabled in config or this environment.", success=False)
if target not in ("memory", "user"):
return json.dumps({"success": False, "error": f"Invalid target '{target}'. Use 'memory' or 'user'."}, ensure_ascii=False)
return tool_error(f"Invalid target '{target}'. Use 'memory' or 'user'.", success=False)
if action == "add":
if not content:
return json.dumps({"success": False, "error": "Content is required for 'add' action."}, ensure_ascii=False)
return tool_error("Content is required for 'add' action.", success=False)
result = store.add(target, content)
elif action == "replace":
if not old_text:
return json.dumps({"success": False, "error": "old_text is required for 'replace' action."}, ensure_ascii=False)
return tool_error("old_text is required for 'replace' action.", success=False)
if not content:
return json.dumps({"success": False, "error": "content is required for 'replace' action."}, ensure_ascii=False)
return tool_error("content is required for 'replace' action.", success=False)
result = store.replace(target, old_text, content)
elif action == "remove":
if not old_text:
return json.dumps({"success": False, "error": "old_text is required for 'remove' action."}, ensure_ascii=False)
return tool_error("old_text is required for 'remove' action.", success=False)
result = store.remove(target, old_text)
else:
return json.dumps({"success": False, "error": f"Unknown action '{action}'. Use: add, replace, remove"}, ensure_ascii=False)
return tool_error(f"Unknown action '{action}'. Use: add, replace, remove", success=False)
return json.dumps(result, ensure_ascii=False)
@@ -539,7 +547,7 @@ MEMORY_SCHEMA = {
# --- Registry ---
from tools.registry import registry
from tools.registry import registry, tool_error
registry.register(
name="memory",

178
tools/warm_session_tool.py Normal file
View File

@@ -0,0 +1,178 @@
"""Warm Session Tool — manage pre-proficient agent sessions.
Allows the agent to build, save, list, and load warm session templates
that pre-seed new sessions with successful tool-call patterns.
"""
import json
import logging
from typing import Optional
from tools.registry import registry
logger = logging.getLogger(__name__)
def warm_session(
action: str,
name: str = None,
description: str = "",
min_messages: int = 20,
max_sessions: int = 20,
source_filter: str = None,
tags: list = None,
) -> str:
"""Manage warm session templates for pre-proficient agent sessions.
Actions:
build — mine existing sessions and create a template
list — show saved templates
load — return a template's conversation_history for injection
delete — remove a template
"""
from agent.warm_session import (
build_from_session_db,
load_template,
list_templates,
build_warm_conversation,
save_template,
TEMPLATES_DIR,
)
if action == "list":
templates = list_templates()
return json.dumps({
"success": True,
"templates": templates,
"count": len(templates),
})
if action == "build":
if not name:
return json.dumps({"success": False, "error": "name is required for 'build'."})
try:
from hermes_state import SessionDB
db = SessionDB()
except Exception as e:
return json.dumps({"success": False, "error": f"Cannot open session DB: {e}"})
template = build_from_session_db(
db,
name=name,
description=description,
min_messages=min_messages,
max_sessions=max_sessions,
source_filter=source_filter,
tags=tags or [],
)
return json.dumps({
"success": True,
"name": template.name,
"example_count": len(template.examples),
"description": template.description,
})
if action == "load":
if not name:
return json.dumps({"success": False, "error": "name is required for 'load'."})
template = load_template(name)
if not template:
return json.dumps({"success": False, "error": f"Template '{name}' not found."})
conversation = build_warm_conversation(template)
return json.dumps({
"success": True,
"name": template.name,
"message_count": len(conversation),
"conversation_preview": [
{"role": m["role"], "content_preview": str(m.get("content", ""))[:100]}
for m in conversation[:6]
],
})
if action == "delete":
if not name:
return json.dumps({"success": False, "error": "name is required for 'delete'."})
path = TEMPLATES_DIR / f"{name}.json"
if not path.exists():
return json.dumps({"success": False, "error": f"Template '{name}' not found."})
path.unlink()
return json.dumps({"success": True, "message": f"Template '{name}' deleted."})
return json.dumps({
"success": False,
"error": f"Unknown action '{action}'. Use: build, list, load, delete",
})
WARM_SESSION_SCHEMA = {
"name": "warm_session",
"description": (
"Manage warm session templates for pre-proficient agent sessions. "
"Marathon sessions have lower error rates than mid-length ones because "
"agents accumulate successful patterns. Warm templates capture those "
"patterns and pre-seed new sessions with experience.\n\n"
"Actions:\n"
" build — mine existing sessions for successful tool-call patterns, save as template\n"
" list — show saved templates\n"
" load — retrieve a template's conversation history for session injection\n"
" delete — remove a template"
),
"parameters": {
"type": "object",
"properties": {
"action": {
"type": "string",
"enum": ["build", "list", "load", "delete"],
"description": "The action to perform.",
},
"name": {
"type": "string",
"description": "Template name. Required for build/load/delete.",
},
"description": {
"type": "string",
"description": "Description for the template. Used with 'build'.",
},
"min_messages": {
"type": "integer",
"description": "Minimum message count to consider a session experienced (default: 20).",
},
"max_sessions": {
"type": "integer",
"description": "Maximum sessions to scan when building (default: 20).",
},
"source_filter": {
"type": "string",
"description": "Filter sessions by source (cli, telegram, discord, etc.).",
},
"tags": {
"type": "array",
"items": {"type": "string"},
"description": "Tags for organizing templates.",
},
},
"required": ["action"],
},
}
registry.register(
name="warm_session",
toolset="skills",
schema=WARM_SESSION_SCHEMA,
handler=lambda args, **kw: warm_session(
action=args.get("action", ""),
name=args.get("name"),
description=args.get("description", ""),
min_messages=args.get("min_messages", 20),
max_sessions=args.get("max_sessions", 20),
source_filter=args.get("source_filter"),
tags=args.get("tags"),
),
emoji="🔥",
)