Skip to content

Hello-QM/catgo-LRG / server/catgo/workflow

catgo-workflow

Verified

Create and manage computational chemistry workflows with CatGo. Supports VASP, CP2K, ORCA, MLP, LAMMPS. Build OER/HER/CO2RR workflows, geometry optimization, frequency analysis, Gibbs energy calculations.

FileKindSize
SKILL.mdskill md3.8 KB
__init__.pyscript1.1 KB
builtins.pyscript4.7 KB
builtins_impl.pyscript21.2 KB
config.pyscript6.5 KB
db.pyscript25.3 KB
engine/__init__.pyscript615 B
engine/advancer.pyscript9.3 KB
engine/ai_diagnosis.pyscript3.8 KB
engine/batch_submitter.pyscript9.7 KB
engine/broadcast.pyscript1.8 KB
engine/collector.pyscript8.3 KB
engine/control_flow.pyscript5.7 KB
engine/engine_builtins.pyscript19.6 KB
engine/engine_registry.pyscript1.5 KB
engine/error_handler.pyscript3.6 KB
engine/hpc_utils.pyscript4.0 KB
engine/job_script.pyscript11.6 KB
engine/lifecycle.pyscript10.8 KB
engine/orca_progress.pyscript3.4 KB
engine/poller.pyscript11.7 KB
engine/resolver.pyscript2.7 KB
engine/result_collector.pyscript56.3 KB
engine/result_handler.pyscript5.6 KB
engine/scanner.pyscript52.0 KB
engine/smart_recovery.pyscript3.6 KB
engine/submitter.pyscript26.2 KB
engine/v1_monitor.pyscript2.7 KB
engine/vasp_submission.pyscript29.0 KB
graph_converter.pyscript6.5 KB
mcp_tools.pyscript7.5 KB
provenance.pyscript5.8 KB
reference.pyscript1.1 KB
service.pyscript4.0 KB
state_map.pyscript1.5 KB
states.pyscript7.5 KB
task_decorator.pyscript2.5 KB
task_ids.pyscript986 B
v1_compat.pyscript2.8 KB
workflow.pyscript12.0 KB

engine/ai_diagnosis.py

View on GitHub
"""AI-assisted error diagnosis for failed HPC tasks.

Reads error logs from HPC and formats structured diagnosis that can be:
1. Automatically processed by the error handler (if confidence is high)
2. Returned via MCP tool for manual AI agent review
"""
from __future__ import annotations

import json
import logging
import re
from typing import Any

from catgo.workflow.db import WorkflowDB

logger = logging.getLogger(__name__)


async def read_error_log(hpc, task: dict, max_lines: int = 200) -> str:
    """Read error log from HPC for a failed task."""
    work_dir = task.get("work_dir")
    if not work_dir or not hpc:
        return ""

    software = task.get("software", "vasp")
    # Try different log files based on software
    log_files = {
        "vasp": ["OUTCAR", "vasp.out", "std_err.txt", "custodian_run.log"],
        "orca": ["ORCA.out", "*.err"],
        "cp2k": ["cp2k.out", "*.err"],
    }
    files_to_check = log_files.get(software, ["*.out", "*.err"])

    for filename in files_to_check:
        try:
            result = await hpc.run_on_owner(lambda filename=filename: hpc.conn.run(
                f"tail -n {max_lines} {work_dir}/{filename} 2>/dev/null",
                check=False,
            ))
            if result.exit_status == 0 and result.stdout.strip():
                return result.stdout.strip()
        except Exception:
            continue
    return ""


def format_diagnosis_prompt(task: dict, error_log: str) -> str:
    """Format a structured prompt for AI diagnosis."""
    params = json.loads(task.get("params_json", "{}") or "{}")
    return (
        f"Analyze this {task.get('software', 'unknown')} calculation error "
        f"and suggest specific parameter fixes.\n\n"
        f"Task type: {task.get('task_type', 'unknown')}\n"
        f"Software: {task.get('software', 'unknown')}\n"
        f"Current parameters: {json.dumps(params, indent=2)}\n"
        f"Error message: {task.get('error_message', 'N/A')}\n\n"
        f"Error log (last lines):\n{error_log[-3000:]}\n\n"
        'Respond with:\n'
        '1. Root cause (one sentence)\n'
        '2. Suggested parameter changes as JSON: {"param": "value", ...}\n'
        '3. Confidence (high/medium/low)'
    )


def parse_diagnosis_response(response: str) -> dict | None:
    """Parse AI response for parameter fixes."""
    json_match = re.search(r'\{[^{}]+\}', response)
    if json_match:
        try:
            fixes = json.loads(json_match.group())
            confidence = "medium"
            if "high" in response.lower():
                confidence = "high"
            elif "low" in response.lower():
                confidence = "low"
            return {
                "fixes": fixes,
                "confidence": confidence,
                "diagnosis": response[:500],
                "tier": 2.5,
            }
        except json.JSONDecodeError:
            pass
    return None


async def get_diagnosis_for_mcp(db: WorkflowDB, task_id: str) -> dict:
    """Get structured error diagnosis for MCP tool response."""
    task = db.get_task(task_id)
    error_msg = task.get("error_message", "No error message")
    params = json.loads(task.get("params_json", "{}") or "{}")

    # Get rule-based diagnosis first
    from catgo.workflow.engine.smart_recovery import diagnose_and_fix

    rule_fix = diagnose_and_fix(db, task, {})

    return {
        "task_id": task_id,
        "task_type": task.get("task_type"),
        "software": task.get("software"),
        "status": task.get("status"),
        "error_message": error_msg,
        "work_dir": task.get("work_dir"),
        "current_params": params,
        "rule_based_diagnosis": rule_fix,
        "hint": (
            "Use catgo_workflow_engine(action='modify_params') to apply fixes, "
            "then catgo_workflow_engine(action='retry') to retry."
        ),
    }
planning-and-task-breakdownaddyosmani/agent-skillsBreaks work into ordered tasks. Use when you have a spec or clear requirements and need to break work into implementable tasks. Use when a task feels too large to start, when you need to estimate scope, or when parallel work is possible.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
screenpipethedotmack/claude-memSearch screen and audio history recorded by Screenpipe and optionally save selected context into claude-mem. Use when the user asks to recall desktop activity, a meeting, or something seen in another app through Screenpipe.Stars: 118kUpdated Claude CodeCodexVerifiedInstall
contact-widgetnexu-io/open-designSelf-contained floating chat widget with welcome screen, social links, meeting button, and message input. Single HTML file, zero dependencies.Stars: 108kUpdated Claude CodeCodexVerifiedInstall
meeting-notesnexu-io/open-designMeeting notes page — title bar with attendees, agenda checklist, decisions block, action items table with owners + dates, and a "next meeting" footer. Use when the brief mentions "meeting notes", "minutes", "1:1 notes", "all-hands recap", or "会议纪要".Stars: 108kUpdated Claude CodeCodexVerifiedInstall