Spaces:

gesh75
/

multivendor-cli-configurator

Running

multivendor-cli-configurator / scripts /parse.py

Upload folder using huggingface_hub

1bd64c1 verified 9 days ago

21.7 kB

#!/usr/bin/env python3
"""
Parse community-sourced Cisco markdown cheatsheets into commands.json.
Schema: { os, role, vendor, cat, title, cmd, desc }
"""
import json, re, hashlib, pathlib, sys

SRC_DIR = pathlib.Path(__file__).parent / "sources"
OUT_FILE = pathlib.Path(__file__).parent.parent / "commands.json"

# Heuristic category map. Order matters: first match wins.
CAT_KEYWORDS = [
    ("BGP",         ["bgp", "ebgp", "ibgp", "as-path", "route-map.*bgp", "neighbor.*remote-as"]),
    ("OSPF",        ["ospf", "ospfv3", "router-id", "lsa", "designated router", "passive-interface"]),
    ("EIGRP",       ["eigrp", "feasible", "successor"]),
    ("Static",      ["static route", "ip route ", "ipv6 route", "default route", "default gateway"]),
    ("VLAN",        ["vlan", "trunk", "access vlan", "voice vlan", "dot1q", "vtp ", "switchport"]),
    ("Spanning-Tree", ["spanning-tree", "stp ", "rapid-pvst", "mst ", "portfast", "bpduguard", "rstp"]),
    ("EtherChannel",["etherchannel", "port-channel", "channel-group", "lacp", "pagp"]),
    ("Interfaces",  ["interface ", "no shutdown", "shutdown", "description", "mtu ", "speed ", "duplex", "loopback"]),
    ("NAT",         ["nat ", "pat ", "ip nat ", "overload", "xlate"]),
    ("ACL",         ["access-list", "access-group", "permit ", "deny ", "ip access-list"]),
    ("DHCP",        ["dhcp", "ip helper", "dhcp pool"]),
    ("HA",          ["hsrp", "vrrp", "glbp", "standby ", "failover"]),
    ("VPN",         ["crypto ", "ipsec", "ike", "isakmp", "vpn", "tunnel"]),
    ("AAA",         ["aaa ", "tacacs", "radius", "authentication", "authorization"]),
    ("SNMP",        ["snmp-server", "snmp "]),
    ("NTP",         ["ntp ", "clock"]),
    ("Logging",     ["logging ", "syslog"]),
    ("QoS",         ["class-map", "policy-map", "service-policy", "qos", "dscp", "cos "]),
    ("Multicast",   ["multicast", "igmp", "pim ", "msdp"]),
    ("MPLS",        ["mpls", "ldp", "vrf ", "rd ", "route-target"]),
    ("Security",    ["port-security", "dhcp snooping", "arp inspection", "storm-control"]),
    ("Troubleshooting", ["ping ", "traceroute", "debug ", "show tech", "show log", "monitor capture"]),
    ("Routing",     ["show ip route", "show route", "ip routing"]),
    ("System",      ["hostname", "show version", "show running", "copy running", "reload", "show inventory", "show env", "show cpu", "show mem"]),
]

# OS inference heuristic
def detect_os(cmd: str, hint: str = "", src: str = "") -> str:
    s = (src or "").lower()
    # Strongest signal: source filename
    if "junos" in s or "juniper" in s or "srx" in s or "mx-" in s:
        return "junos"
    if "arista" in s or "/eos" in s or "alista" in s:
        return "eos"
    if "nxos" in s or "nexus" in s or "nx-os" in s or "nx9k" in s or "9k" in s:
        return "nxos"
    if "asa" in s or "firepower" in s or "ftd" in s:
        return "asa"
    c = (cmd + " " + hint).lower()
    if any(k in c for k in ["set interfaces", "set protocols", "set system", "show configuration |", "commit confirmed", "set routing-options", "set security zones"]):
        return "junos"
    first_word = c.split("\n", 1)[0].split(" ", 1)[0] if c else ""
    if first_word == "feature" or " feature vpc" in c or " feature ospf" in c:
        return "nxos"
    if "no switchport" in c and "/" in c and "ip address" in c:
        return "nxos"
    if "nameif" in c or "security-level " in c or "packet-tracer input" in c:
        return "asa"
    if "configure session" in c or "ip virtual-router" in c or "mlag configuration" in c or "service routing protocols model" in c:
        return "eos"
    return "ios"

# Role inference
def detect_role(cmd: str, sec_path: str) -> str:
    c = (cmd + " " + sec_path).lower()
    if any(k in c for k in ["zone", "policy.*permit", "nameif", "security-level", "vpn", "ipsec", "ike", "failover"]):
        return "firewall"
    if any(k in c for k in ["vlan ", "switchport", "spanning-tree", "trunk", "vtp ", "port-channel"]):
        return "switch"
    return "router"

# Vendor inference (we're mostly Cisco from these sources, but be safe)
def detect_vendor(cmd: str, source_name: str) -> str:
    c = cmd.lower()
    if "set " in c and any(k in c for k in ["interfaces ge-", "protocols ospf", "routing-options"]):
        return "Juniper"
    if "configure session" in c or "ip virtual-router" in c:
        return "Arista"
    return "Cisco"

def categorize(cmd: str, section: str) -> str:
    text = (cmd + " " + section).lower()
    for cat, keys in CAT_KEYWORDS:
        for k in keys:
            if re.search(k, text):
                return cat
    return "System"

def clean(s: str) -> str:
    # strip prompts, leading "$ " or "# " from shell, leading bullets
    s = re.sub(r"^[\s]*[$#>][\s]?", "", s, flags=re.MULTILINE)
    s = re.sub(r"^\s*[-*+]\s+", "", s, flags=re.MULTILINE)
    return s.rstrip()

def short_title(cmd: str, fallback: str) -> str:
    first = cmd.strip().split("\n")[0]
    first = re.sub(r"\s+#.*$", "", first)  # strip inline comment
    if len(first) > 70: first = first[:67] + "..."
    return first or fallback or "command"

# headings: track h2/h3/h4/h5 path
def parse_markdown(path: pathlib.Path):
    text = path.read_text(encoding="utf-8", errors="ignore")
    lines = text.split("\n")
    headings = {}  # level -> title
    items = []
    i = 0
    n = len(lines)
    while i < n:
        line = lines[i]
        m = re.match(r"^(#{2,6})\s+(.+?)\s*$", line)
        if m:
            lvl = len(m.group(1))
            title = re.sub(r"[`*_:🌐📘🚀📑]", "", m.group(2)).strip()
            # collapse emoji/special chars
            title = re.sub(r"[^\w\s\-\(\)\/.,]", "", title).strip()
            headings[lvl] = title
            # clear deeper levels
            for k in list(headings):
                if k > lvl: del headings[k]
            i += 1
            continue
        # detect fenced code blocks
        m2 = re.match(r"^```(\w*)\s*$", line)
        if m2:
            lang = m2.group(1).lower()
            block = []
            i += 1
            while i < n and not lines[i].startswith("```"):
                block.append(lines[i])
                i += 1
            i += 1  # skip closing ```
            # Only keep blocks that look like config (filter out python/bash unless small)
            if lang in {"python", "json", "yaml", "javascript", "js", "html"}:
                continue
            code = clean("\n".join(block)).strip()
            if not code: continue
            if len(code) > 1500: continue  # skip huge dumps
            # Split very long blocks into sub-commands by blank lines or by ! separators if it's clearly multiple
            chunks = [c.strip() for c in re.split(r"\n\s*\n", code) if c.strip()]
            for chunk in chunks:
                # Skip pure prose blocks
                if not any(re.search(r"\b(show|interface|ip|router|vlan|set|access-list|hostname|configure|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute)\b", l, re.I) for l in chunk.split("\n")):
                    continue
                section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                items.append({
                    "src": path.name,
                    "section": section,
                    "cmd": chunk,
                })
            continue
        # Markdown table: detect "Command|Description" or "Commands|...|" header followed by separator
        if "|" in line and i+1 < n and re.match(r"^\s*[-:\s|]+\s*$", lines[i+1]) and any(h in line.lower() for h in ["command", "syntax", "snippet"]):
            i += 2  # skip header + separator
            while i < n and "|" in lines[i] and lines[i].strip():
                row = lines[i].split("|")
                if len(row) >= 2:
                    raw_cmd = row[0].strip()
                    desc = row[1].strip() if len(row) >= 2 else ""
                    # strip backticks/prompts
                    cmd = re.sub(r"^`+|`+$", "", raw_cmd)
                    cmd = re.sub(r"^[Rr]\d+(\(config[^)]*\))?[#>]\s*", "", cmd)
                    cmd = re.sub(r"^[Ss]witch(\(config[^)]*\))?[#>]\s*", "", cmd)
                    cmd = re.sub(r"^[Hh]ost(\(config[^)]*\))?[#>]\s*", "", cmd)
                    cmd = cmd.strip()
                    desc = re.sub(r"`([^`]+)`", r"\1", desc)  # remove inline backticks in desc
                    if cmd and len(cmd) < 250 and re.search(r"\b(show|interface|ip|ipv6|router|vlan|set|access-list|hostname|configure|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute|no |service|line|standby|vrrp|hsrp)\b", cmd, re.I):
                        section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                        items.append({"src": path.name, "section": section + " | " + (desc or ""), "cmd": cmd})
                i += 1
            continue
        # Blockquote with inline backticks: > ``R1(config)#ip route ...``
        bq = re.match(r"^>\s*`{1,2}([^`]+)`{1,2}\s*$", line)
        if bq:
            raw_cmd = bq.group(1).strip()
            cmd = strip_prompt(raw_cmd)
            if cmd and re.search(r"\b(show|interface|ip|ipv6|router|vlan|set|access-list|hostname|configure|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute|no )\b", cmd, re.I):
                section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                items.append({"src": path.name, "section": section, "cmd": cmd})
            i += 1
            continue
        # AsciiDoc table: [cols=...] ... |=== |Cmd| Desc |=== — rows are alternating |cell lines
        if line.strip() == "|===":
            i += 1
            # consume table until next |===
            rows = []
            cur_row = []
            while i < n and lines[i].strip() != "|===":
                ln = lines[i]
                m_cell = re.match(r"^\|\s?(.*)$", ln)
                if m_cell:
                    if cur_row and (len(cur_row) >= 2 or ln.startswith("|")):
                        # new cell starts → row of 2 cells = command + desc
                        if len(cur_row) >= 2:
                            rows.append(cur_row)
                            cur_row = []
                    cur_row.append(m_cell.group(1))
                else:
                    # continuation of last cell
                    if cur_row:
                        cur_row[-1] += "\n" + ln
                i += 1
            if cur_row and len(cur_row) >= 2:
                rows.append(cur_row)
            i += 1  # skip closing |===
            # skip first row if it looks like header (Command/Description)
            for r in rows:
                raw_cmd, desc = r[0].strip(), r[1].strip() if len(r) > 1 else ""
                if re.match(r"^\s*command\s*$", raw_cmd, re.I): continue
                if re.match(r"^\s*descritp?tion\s*$", desc, re.I): continue
                # strip asciidoc emphasis (*x* or _x_) and backticks
                cmd = re.sub(r"[*`_]", "", raw_cmd).strip()
                cmd = strip_prompt(cmd)
                if cmd and len(cmd) < 300 and re.search(r"\b(show|interface|ip|ipv6|router|vlan|set|access-list|hostname|configure|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute|no |request|commit|delete|run)\b", cmd, re.I):
                    section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                    items.append({"src": path.name, "section": section + " | " + desc[:120], "cmd": cmd})
            continue
        # Plain prompt lines anywhere: "Switch# show ...", "Router(config)# ip route ...", "user@JunOS> show ..."
        pm = re.match(r"^\s*(?:user@\w+|R\d+|Switch|Router|Host|sw\d+|[A-Z][\w-]*?)(\(config[^)]*\))?[#>%]\s+(.+?)\s*$", line)
        if pm and len(pm.group(2)) > 3 and len(pm.group(2)) < 200:
            cmd = pm.group(2).strip()
            # filter out non-config lines (output text)
            if re.search(r"^\s*(?:show|interface|ip |ipv6|router|vlan|set |access-list|hostname|configure|conf t|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute|no |request|commit|delete|run|edit|exit|end|description|mtu|speed|duplex|shutdown|sh )", cmd, re.I):
                section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                items.append({"src": path.name, "section": section, "cmd": cmd})
        # Indented (4-space) code lines that look like commands
        if line.startswith("    ") and not line.strip().startswith(("//", "#", "--", "|", "*", "-")):
            inner = line.strip()
            # match: optional `quoted` cmd or prompt+cmd or bare command
            inner = re.sub(r'^"([^"]+)"\s*(#.*)?$', r"\1", inner)
            cmd = strip_prompt(inner)
            cmd = cmd.split("#")[0].strip()  # strip inline cisco-style comment
            if cmd and 6 <= len(cmd) <= 200 and re.match(r"^(?:show|interface|ip|ipv6|router|vlan|set|access-list|hostname|configure|switchport|spanning-tree|crypto|snmp|logging|ntp|aaa|debug|enable|copy|write|reload|monitor|clear|ping|traceroute|no |service|request|commit|delete|edit|description)\b", cmd, re.I):
                section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
                items.append({"src": path.name, "section": section, "cmd": cmd})
        i += 1
    return items

def strip_prompt(s: str) -> str:
    s = re.sub(r"^(?:user@\w+|R\d+|Switch|Router|Host|sw\d+|S\d+|[A-Z][\w-]*?)(\(config[^)]*\))?[#>%]\s*", "", s)
    return s.strip()

def main():
    raw = []
    for md in sorted(SRC_DIR.glob("*.md")):
        # ENCOR/ENARSI has its own dedicated parser → encor.json
        if "encor" in md.name.lower(): continue
        # Junos books/reference have their own dedicated parser → junos.json
        if md.name.startswith("junos-dayone") or md.name == "junos-os-cli-reference.md": continue
        # Arista EOS user guide has its own dedicated parser → arista.json
        if md.name.startswith("arista-eos"): continue
        # Cisco IOS OSPF reference has its own dedicated parser → cisco_ospf.json
        if md.name == "cisco-ios-ospf-reference.md": continue
        # Cisco ASA 9.24 + NX-OS VXLAN/BGP-EVPN have their own dedicated parsers
        if md.name in {"cisco-asa-924-reference.md", "cisco-nxos-vxlan-evpn-reference.md"}: continue
        try:
            items = parse_markdown(md)
            raw.extend(items)
            print(f"  parsed {md.name}: {len(items)} blocks", file=sys.stderr)
        except Exception as e:
            print(f"  ! {md.name}: {e}", file=sys.stderr)

    # Convert to schema
    cmds = []
    seen = set()
    for it in raw:
        cmd = it["cmd"]
        sec = it["section"]
        # dedup key on normalized command
        key = re.sub(r"\s+", " ", cmd).lower()
        if key in seen: continue
        seen.add(key)
        os = detect_os(cmd, sec, it["src"])
        role = detect_role(cmd, sec)
        vendor = detect_vendor(cmd, it["src"])
        # vendor override from filename
        sf = it["src"].lower()
        if "arista" in sf or "alista" in sf or "eos" in sf: vendor = "Arista"
        elif "junos" in sf or "juniper" in sf or "srx" in sf: vendor = "Juniper"
        cat = categorize(cmd, sec)
        title = short_title(cmd, sec.split(" > ")[-1] if sec else "")
        cmds.append({
            "os": os, "role": role, "vendor": vendor, "cat": cat,
            "title": title, "cmd": cmd,
            "desc": sec or "",
        })

    # Merge with the hand-curated dataset extracted from cheatsheet.html
    seed_path = pathlib.Path(__file__).parent / "seed.json"
    if seed_path.exists():
        seed = json.loads(seed_path.read_text())
        for s in seed:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s)
        print(f"  merged seed: {len(seed)} curated entries", file=sys.stderr)

    # Merge ENCOR/ENARSI Cisco Press dataset (1,391 commands)
    encor_path = pathlib.Path(__file__).parent / "encor.json"
    if encor_path.exists():
        encor = json.loads(encor_path.read_text())
        added = 0
        for s in encor:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged ENCOR: {added} new entries (deduped from {len(encor)})", file=sys.stderr)

    # Merge Junos books (Day One Beginners + Day One CLI + CLI Reference index)
    junos_path = pathlib.Path(__file__).parent / "junos.json"
    if junos_path.exists():
        junos = json.loads(junos_path.read_text())
        added = 0
        for s in junos:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged Junos: {added} new entries (deduped from {len(junos)})", file=sys.stderr)

    # Merge Arista EOS User Guide
    arista_path = pathlib.Path(__file__).parent / "arista.json"
    if arista_path.exists():
        arista = json.loads(arista_path.read_text())
        added = 0
        for s in arista:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged Arista: {added} new entries (deduped from {len(arista)})", file=sys.stderr)

    # Merge Cisco IOS OSPF dedicated reference (111 commands)
    ospf_path = pathlib.Path(__file__).parent / "cisco_ospf.json"
    if ospf_path.exists():
        ospf_set = json.loads(ospf_path.read_text())
        added = 0
        for s in ospf_set:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged Cisco OSPF: {added} new entries (deduped from {len(ospf_set)})", file=sys.stderr)

    # Merge Cisco ASA 9.24 + NX-OS VXLAN BGP EVPN
    for extra_name in ("cisco_asa.json", "cisco_nxos_vxlan.json"):
        p = pathlib.Path(__file__).parent / extra_name
        if not p.exists(): continue
        extras = json.loads(p.read_text())
        added = 0
        for s in extras:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged {extra_name}: {added} new entries (deduped from {len(extras)})", file=sys.stderr)

    # Merge FRR (FRRouting) dataset — derived from the official FRR docs
    # via scripts/parse_frr.py. Source JSON lives outside the repo at
    # 06_Documentation/FRR_CLI_Command_Reference.json; the parser checks it in
    # as scripts/frr.json so this merger stays purely local.
    frr_path = pathlib.Path(__file__).parent / "frr.json"
    if frr_path.exists():
        frr = json.loads(frr_path.read_text())
        added = 0
        for s in frr:
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key not in seen:
                seen.add(key); cmds.append(s); added += 1
        print(f"  merged FRR: {added} new entries (deduped from {len(frr)})", file=sys.stderr)

    # Merge "wave 2" datasets emitted by scripts/parse_extras.py — 7 new
    # vendors (VyOS, SONiC, NVIDIA NVUE, Palo Alto PAN-OS, Nokia SR Linux,
    # FortiOS, Mikrotik RouterOS) and 2 backfill files for Cisco / Juniper
    # gaps. All share the same row schema so we use a uniform dedupe loop.
    extras = [
        "cisco_gaps.json", "juniper_gaps.json",
        "vyos.json", "sonic.json", "nvue.json", "panos.json",
        "srlinux.json", "fortios.json", "routeros.json", "exos.json",
        "aoscx.json", "vrp.json",
    ]
    for name in extras:
        p = pathlib.Path(__file__).parent / name
        if not p.exists():
            continue
        rows = json.loads(p.read_text())
        added = 0
        for s in rows:
            if not (s.get("cmd") and s.get("vendor")):
                continue
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key in seen:
                continue
            seen.add(key); cmds.append(s); added += 1
        print(f"  merged {name}: {added} new entries (deduped from {len(rows)})", file=sys.stderr)

    # Merge external dataset (~10k commands previously curated outside this repo).
    # Already in the same {os,role,vendor,cat,title,cmd,desc} schema → straight dedupe by cmd.
    ext_path = pathlib.Path(__file__).parent / "external_merged.json"
    if ext_path.exists():
        ext = json.loads(ext_path.read_text())
        added = 0
        for s in ext:
            if not (s.get("cmd") and s.get("vendor")):
                continue
            key = re.sub(r"\s+", " ", s["cmd"]).lower()
            if key in seen:
                continue
            seen.add(key); cmds.append(s); added += 1
        print(f"  merged external_merged: {added} new entries (deduped from {len(ext)})", file=sys.stderr)

    OUT_FILE.write_text(json.dumps(cmds, indent=1, ensure_ascii=False))
    print(f"\nwrote {len(cmds)} commands -> {OUT_FILE}")
    # quick stats
    from collections import Counter
    print("\nBy OS:", dict(Counter(c["os"] for c in cmds)))
    print("By role:", dict(Counter(c["role"] for c in cmds)))
    print("By cat:", dict(Counter(c["cat"] for c in cmds)))

if __name__ == "__main__":
    main()

	#!/usr/bin/env python3
	"""
	Parse community-sourced Cisco markdown cheatsheets into commands.json.
	Schema: { os, role, vendor, cat, title, cmd, desc }
	"""
	import json, re, hashlib, pathlib, sys

	SRC_DIR = pathlib.Path(__file__).parent / "sources"
	OUT_FILE = pathlib.Path(__file__).parent.parent / "commands.json"

	# Heuristic category map. Order matters: first match wins.
	CAT_KEYWORDS = [
	("BGP", ["bgp", "ebgp", "ibgp", "as-path", "route-map.bgp", "neighbor.remote-as"]),
	("OSPF", ["ospf", "ospfv3", "router-id", "lsa", "designated router", "passive-interface"]),
	("EIGRP", ["eigrp", "feasible", "successor"]),
	("Static", ["static route", "ip route ", "ipv6 route", "default route", "default gateway"]),
	("VLAN", ["vlan", "trunk", "access vlan", "voice vlan", "dot1q", "vtp ", "switchport"]),
	("Spanning-Tree", ["spanning-tree", "stp ", "rapid-pvst", "mst ", "portfast", "bpduguard", "rstp"]),
	("EtherChannel",["etherchannel", "port-channel", "channel-group", "lacp", "pagp"]),
	("Interfaces", ["interface ", "no shutdown", "shutdown", "description", "mtu ", "speed ", "duplex", "loopback"]),
	("NAT", ["nat ", "pat ", "ip nat ", "overload", "xlate"]),
	("ACL", ["access-list", "access-group", "permit ", "deny ", "ip access-list"]),
	("DHCP", ["dhcp", "ip helper", "dhcp pool"]),
	("HA", ["hsrp", "vrrp", "glbp", "standby ", "failover"]),
	("VPN", ["crypto ", "ipsec", "ike", "isakmp", "vpn", "tunnel"]),
	("AAA", ["aaa ", "tacacs", "radius", "authentication", "authorization"]),
	("SNMP", ["snmp-server", "snmp "]),
	("NTP", ["ntp ", "clock"]),
	("Logging", ["logging ", "syslog"]),
	("QoS", ["class-map", "policy-map", "service-policy", "qos", "dscp", "cos "]),
	("Multicast", ["multicast", "igmp", "pim ", "msdp"]),
	("MPLS", ["mpls", "ldp", "vrf ", "rd ", "route-target"]),
	("Security", ["port-security", "dhcp snooping", "arp inspection", "storm-control"]),
	("Troubleshooting", ["ping ", "traceroute", "debug ", "show tech", "show log", "monitor capture"]),
	("Routing", ["show ip route", "show route", "ip routing"]),
	("System", ["hostname", "show version", "show running", "copy running", "reload", "show inventory", "show env", "show cpu", "show mem"]),
	]

	# OS inference heuristic
	def detect_os(cmd: str, hint: str = "", src: str = "") -> str:
	s = (src or "").lower()
	# Strongest signal: source filename
	if "junos" in s or "juniper" in s or "srx" in s or "mx-" in s:
	return "junos"
	if "arista" in s or "/eos" in s or "alista" in s:
	return "eos"
	if "nxos" in s or "nexus" in s or "nx-os" in s or "nx9k" in s or "9k" in s:
	return "nxos"
	if "asa" in s or "firepower" in s or "ftd" in s:
	return "asa"
	c = (cmd + " " + hint).lower()
	if any(k in c for k in ["set interfaces", "set protocols", "set system", "show configuration \|", "commit confirmed", "set routing-options", "set security zones"]):
	return "junos"
	first_word = c.split("\n", 1)[0].split(" ", 1)[0] if c else ""
	if first_word == "feature" or " feature vpc" in c or " feature ospf" in c:
	return "nxos"
	if "no switchport" in c and "/" in c and "ip address" in c:
	return "nxos"
	if "nameif" in c or "security-level " in c or "packet-tracer input" in c:
	return "asa"
	if "configure session" in c or "ip virtual-router" in c or "mlag configuration" in c or "service routing protocols model" in c:
	return "eos"
	return "ios"

	# Role inference
	def detect_role(cmd: str, sec_path: str) -> str:
	c = (cmd + " " + sec_path).lower()
	if any(k in c for k in ["zone", "policy.*permit", "nameif", "security-level", "vpn", "ipsec", "ike", "failover"]):
	return "firewall"
	if any(k in c for k in ["vlan ", "switchport", "spanning-tree", "trunk", "vtp ", "port-channel"]):
	return "switch"
	return "router"

	# Vendor inference (we're mostly Cisco from these sources, but be safe)
	def detect_vendor(cmd: str, source_name: str) -> str:
	c = cmd.lower()
	if "set " in c and any(k in c for k in ["interfaces ge-", "protocols ospf", "routing-options"]):
	return "Juniper"
	if "configure session" in c or "ip virtual-router" in c:
	return "Arista"
	return "Cisco"

	def categorize(cmd: str, section: str) -> str:
	text = (cmd + " " + section).lower()
	for cat, keys in CAT_KEYWORDS:
	for k in keys:
	if re.search(k, text):
	return cat
	return "System"

	def clean(s: str) -> str:
	# strip prompts, leading "$ " or "# " from shell, leading bullets
	s = re.sub(r"^[\s]*[$#>][\s]?", "", s, flags=re.MULTILINE)
	s = re.sub(r"^\s[-+]\s+", "", s, flags=re.MULTILINE)
	return s.rstrip()

	def short_title(cmd: str, fallback: str) -> str:
	first = cmd.strip().split("\n")[0]
	first = re.sub(r"\s+#.*$", "", first) # strip inline comment
	if len(first) > 70: first = first[:67] + "..."
	return first or fallback or "command"

	# headings: track h2/h3/h4/h5 path
	def parse_markdown(path: pathlib.Path):
	text = path.read_text(encoding="utf-8", errors="ignore")
	lines = text.split("\n")
	headings = {} # level -> title
	items = []
	i = 0
	n = len(lines)
	while i < n:
	line = lines[i]
	m = re.match(r"^(#{2,6})\s+(.+?)\s*$", line)
	if m:
	lvl = len(m.group(1))
	title = re.sub(r"[`*_:🌐📘🚀📑]", "", m.group(2)).strip()
	# collapse emoji/special chars
	title = re.sub(r"[^\w\s\-\(\)\/.,]", "", title).strip()
	headings[lvl] = title
	# clear deeper levels
	for k in list(headings):
	if k > lvl: del headings[k]
	i += 1
	continue
	# detect fenced code blocks
	m2 = re.match(r"^```(\w)\s$", line)
	if m2:
	lang = m2.group(1).lower()
	block = []
	i += 1
	while i < n and not lines[i].startswith("```"):
	block.append(lines[i])
	i += 1
	i += 1 # skip closing ```
	# Only keep blocks that look like config (filter out python/bash unless small)
	if lang in {"python", "json", "yaml", "javascript", "js", "html"}:
	continue
	code = clean("\n".join(block)).strip()
	if not code: continue
	if len(code) > 1500: continue # skip huge dumps
	# Split very long blocks into sub-commands by blank lines or by ! separators if it's clearly multiple
	chunks = [c.strip() for c in re.split(r"\n\s*\n", code) if c.strip()]
	for chunk in chunks:
	# Skip pure prose blocks
	if not any(re.search(r"\b(show\|interface\|ip\|router\|vlan\|set\|access-list\|hostname\|configure\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute)\b", l, re.I) for l in chunk.split("\n")):
	continue
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({
	"src": path.name,
	"section": section,
	"cmd": chunk,
	})
	continue
	# Markdown table: detect "Command\|Description" or "Commands\|...\|" header followed by separator
	if "\|" in line and i+1 < n and re.match(r"^\s[-:\s\|]+\s$", lines[i+1]) and any(h in line.lower() for h in ["command", "syntax", "snippet"]):
	i += 2 # skip header + separator
	while i < n and "\|" in lines[i] and lines[i].strip():
	row = lines[i].split("\|")
	if len(row) >= 2:
	raw_cmd = row[0].strip()
	desc = row[1].strip() if len(row) >= 2 else ""
	# strip backticks/prompts
	cmd = re.sub(r"^`+\|`+$", "", raw_cmd)
	cmd = re.sub(r"^[Rr]\d+(\(config[^)]\))?[#>]\s", "", cmd)
	cmd = re.sub(r"^[Ss]witch(\(config[^)]\))?[#>]\s", "", cmd)
	cmd = re.sub(r"^[Hh]ost(\(config[^)]\))?[#>]\s", "", cmd)
	cmd = cmd.strip()
	desc = re.sub(r"`([^`]+)`", r"\1", desc) # remove inline backticks in desc
	if cmd and len(cmd) < 250 and re.search(r"\b(show\|interface\|ip\|ipv6\|router\|vlan\|set\|access-list\|hostname\|configure\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute\|no \|service\|line\|standby\|vrrp\|hsrp)\b", cmd, re.I):
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({"src": path.name, "section": section + " \| " + (desc or ""), "cmd": cmd})
	i += 1
	continue
	# Blockquote with inline backticks: > ``R1(config)#ip route ...``
	bq = re.match(r"^>\s`{1,2}([^`]+)`{1,2}\s$", line)
	if bq:
	raw_cmd = bq.group(1).strip()
	cmd = strip_prompt(raw_cmd)
	if cmd and re.search(r"\b(show\|interface\|ip\|ipv6\|router\|vlan\|set\|access-list\|hostname\|configure\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute\|no )\b", cmd, re.I):
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({"src": path.name, "section": section, "cmd": cmd})
	i += 1
	continue
	# AsciiDoc table: [cols=...] ... \|=== \|Cmd\| Desc \|=== — rows are alternating \|cell lines
	if line.strip() == "\|===":
	i += 1
	# consume table until next \|===
	rows = []
	cur_row = []
	while i < n and lines[i].strip() != "\|===":
	ln = lines[i]
	m_cell = re.match(r"^\\|\s?(.*)$", ln)
	if m_cell:
	if cur_row and (len(cur_row) >= 2 or ln.startswith("\|")):
	# new cell starts → row of 2 cells = command + desc
	if len(cur_row) >= 2:
	rows.append(cur_row)
	cur_row = []
	cur_row.append(m_cell.group(1))
	else:
	# continuation of last cell
	if cur_row:
	cur_row[-1] += "\n" + ln
	i += 1
	if cur_row and len(cur_row) >= 2:
	rows.append(cur_row)
	i += 1 # skip closing \|===
	# skip first row if it looks like header (Command/Description)
	for r in rows:
	raw_cmd, desc = r[0].strip(), r[1].strip() if len(r) > 1 else ""
	if re.match(r"^\scommand\s$", raw_cmd, re.I): continue
	if re.match(r"^\sdescritp?tion\s$", desc, re.I): continue
	# strip asciidoc emphasis (x or _x_) and backticks
	cmd = re.sub(r"[*`_]", "", raw_cmd).strip()
	cmd = strip_prompt(cmd)
	if cmd and len(cmd) < 300 and re.search(r"\b(show\|interface\|ip\|ipv6\|router\|vlan\|set\|access-list\|hostname\|configure\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute\|no \|request\|commit\|delete\|run)\b", cmd, re.I):
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({"src": path.name, "section": section + " \| " + desc[:120], "cmd": cmd})
	continue
	# Plain prompt lines anywhere: "Switch# show ...", "Router(config)# ip route ...", "user@JunOS> show ..."
	pm = re.match(r"^\s(?:user@\w+\|R\d+\|Switch\|Router\|Host\|sw\d+\|[A-Z][\w-]?)(\(config[^)]\))?[#>%]\s+(.+?)\s$", line)
	if pm and len(pm.group(2)) > 3 and len(pm.group(2)) < 200:
	cmd = pm.group(2).strip()
	# filter out non-config lines (output text)
	if re.search(r"^\s*(?:show\|interface\|ip \|ipv6\|router\|vlan\|set \|access-list\|hostname\|configure\|conf t\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute\|no \|request\|commit\|delete\|run\|edit\|exit\|end\|description\|mtu\|speed\|duplex\|shutdown\|sh )", cmd, re.I):
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({"src": path.name, "section": section, "cmd": cmd})
	# Indented (4-space) code lines that look like commands
	if line.startswith(" ") and not line.strip().startswith(("//", "#", "--", "\|", "*", "-")):
	inner = line.strip()
	# match: optional `quoted` cmd or prompt+cmd or bare command
	inner = re.sub(r'^"([^"]+)"\s(#.)?$', r"\1", inner)
	cmd = strip_prompt(inner)
	cmd = cmd.split("#")[0].strip() # strip inline cisco-style comment
	if cmd and 6 <= len(cmd) <= 200 and re.match(r"^(?:show\|interface\|ip\|ipv6\|router\|vlan\|set\|access-list\|hostname\|configure\|switchport\|spanning-tree\|crypto\|snmp\|logging\|ntp\|aaa\|debug\|enable\|copy\|write\|reload\|monitor\|clear\|ping\|traceroute\|no \|service\|request\|commit\|delete\|edit\|description)\b", cmd, re.I):
	section = " > ".join(headings.get(k, "") for k in sorted(headings) if headings.get(k))
	items.append({"src": path.name, "section": section, "cmd": cmd})
	i += 1
	return items

	def strip_prompt(s: str) -> str:
	s = re.sub(r"^(?:user@\w+\|R\d+\|Switch\|Router\|Host\|sw\d+\|S\d+\|[A-Z][\w-]?)(\(config[^)]\))?[#>%]\s*", "", s)
	return s.strip()

	def main():
	raw = []
	for md in sorted(SRC_DIR.glob("*.md")):
	# ENCOR/ENARSI has its own dedicated parser → encor.json
	if "encor" in md.name.lower(): continue
	# Junos books/reference have their own dedicated parser → junos.json
	if md.name.startswith("junos-dayone") or md.name == "junos-os-cli-reference.md": continue
	# Arista EOS user guide has its own dedicated parser → arista.json
	if md.name.startswith("arista-eos"): continue
	# Cisco IOS OSPF reference has its own dedicated parser → cisco_ospf.json
	if md.name == "cisco-ios-ospf-reference.md": continue
	# Cisco ASA 9.24 + NX-OS VXLAN/BGP-EVPN have their own dedicated parsers
	if md.name in {"cisco-asa-924-reference.md", "cisco-nxos-vxlan-evpn-reference.md"}: continue
	try:
	items = parse_markdown(md)
	raw.extend(items)
	print(f" parsed {md.name}: {len(items)} blocks", file=sys.stderr)
	except Exception as e:
	print(f" ! {md.name}: {e}", file=sys.stderr)

	# Convert to schema
	cmds = []
	seen = set()
	for it in raw:
	cmd = it["cmd"]
	sec = it["section"]
	# dedup key on normalized command
	key = re.sub(r"\s+", " ", cmd).lower()
	if key in seen: continue
	seen.add(key)
	os = detect_os(cmd, sec, it["src"])
	role = detect_role(cmd, sec)
	vendor = detect_vendor(cmd, it["src"])
	# vendor override from filename
	sf = it["src"].lower()
	if "arista" in sf or "alista" in sf or "eos" in sf: vendor = "Arista"
	elif "junos" in sf or "juniper" in sf or "srx" in sf: vendor = "Juniper"
	cat = categorize(cmd, sec)
	title = short_title(cmd, sec.split(" > ")[-1] if sec else "")
	cmds.append({
	"os": os, "role": role, "vendor": vendor, "cat": cat,
	"title": title, "cmd": cmd,
	"desc": sec or "",
	})

	# Merge with the hand-curated dataset extracted from cheatsheet.html
	seed_path = pathlib.Path(__file__).parent / "seed.json"
	if seed_path.exists():
	seed = json.loads(seed_path.read_text())
	for s in seed:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s)
	print(f" merged seed: {len(seed)} curated entries", file=sys.stderr)

	# Merge ENCOR/ENARSI Cisco Press dataset (1,391 commands)
	encor_path = pathlib.Path(__file__).parent / "encor.json"
	if encor_path.exists():
	encor = json.loads(encor_path.read_text())
	added = 0
	for s in encor:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged ENCOR: {added} new entries (deduped from {len(encor)})", file=sys.stderr)

	# Merge Junos books (Day One Beginners + Day One CLI + CLI Reference index)
	junos_path = pathlib.Path(__file__).parent / "junos.json"
	if junos_path.exists():
	junos = json.loads(junos_path.read_text())
	added = 0
	for s in junos:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged Junos: {added} new entries (deduped from {len(junos)})", file=sys.stderr)

	# Merge Arista EOS User Guide
	arista_path = pathlib.Path(__file__).parent / "arista.json"
	if arista_path.exists():
	arista = json.loads(arista_path.read_text())
	added = 0
	for s in arista:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged Arista: {added} new entries (deduped from {len(arista)})", file=sys.stderr)

	# Merge Cisco IOS OSPF dedicated reference (111 commands)
	ospf_path = pathlib.Path(__file__).parent / "cisco_ospf.json"
	if ospf_path.exists():
	ospf_set = json.loads(ospf_path.read_text())
	added = 0
	for s in ospf_set:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged Cisco OSPF: {added} new entries (deduped from {len(ospf_set)})", file=sys.stderr)

	# Merge Cisco ASA 9.24 + NX-OS VXLAN BGP EVPN
	for extra_name in ("cisco_asa.json", "cisco_nxos_vxlan.json"):
	p = pathlib.Path(__file__).parent / extra_name
	if not p.exists(): continue
	extras = json.loads(p.read_text())
	added = 0
	for s in extras:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged {extra_name}: {added} new entries (deduped from {len(extras)})", file=sys.stderr)

	# Merge FRR (FRRouting) dataset — derived from the official FRR docs
	# via scripts/parse_frr.py. Source JSON lives outside the repo at
	# 06_Documentation/FRR_CLI_Command_Reference.json; the parser checks it in
	# as scripts/frr.json so this merger stays purely local.
	frr_path = pathlib.Path(__file__).parent / "frr.json"
	if frr_path.exists():
	frr = json.loads(frr_path.read_text())
	added = 0
	for s in frr:
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key not in seen:
	seen.add(key); cmds.append(s); added += 1
	print(f" merged FRR: {added} new entries (deduped from {len(frr)})", file=sys.stderr)

	# Merge "wave 2" datasets emitted by scripts/parse_extras.py — 7 new
	# vendors (VyOS, SONiC, NVIDIA NVUE, Palo Alto PAN-OS, Nokia SR Linux,
	# FortiOS, Mikrotik RouterOS) and 2 backfill files for Cisco / Juniper
	# gaps. All share the same row schema so we use a uniform dedupe loop.
	extras = [
	"cisco_gaps.json", "juniper_gaps.json",
	"vyos.json", "sonic.json", "nvue.json", "panos.json",
	"srlinux.json", "fortios.json", "routeros.json", "exos.json",
	"aoscx.json", "vrp.json",
	]
	for name in extras:
	p = pathlib.Path(__file__).parent / name
	if not p.exists():
	continue
	rows = json.loads(p.read_text())
	added = 0
	for s in rows:
	if not (s.get("cmd") and s.get("vendor")):
	continue
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key in seen:
	continue
	seen.add(key); cmds.append(s); added += 1
	print(f" merged {name}: {added} new entries (deduped from {len(rows)})", file=sys.stderr)

	# Merge external dataset (~10k commands previously curated outside this repo).
	# Already in the same {os,role,vendor,cat,title,cmd,desc} schema → straight dedupe by cmd.
	ext_path = pathlib.Path(__file__).parent / "external_merged.json"
	if ext_path.exists():
	ext = json.loads(ext_path.read_text())
	added = 0
	for s in ext:
	if not (s.get("cmd") and s.get("vendor")):
	continue
	key = re.sub(r"\s+", " ", s["cmd"]).lower()
	if key in seen:
	continue
	seen.add(key); cmds.append(s); added += 1
	print(f" merged external_merged: {added} new entries (deduped from {len(ext)})", file=sys.stderr)

	OUT_FILE.write_text(json.dumps(cmds, indent=1, ensure_ascii=False))
	print(f"\nwrote {len(cmds)} commands -> {OUT_FILE}")
	# quick stats
	from collections import Counter
	print("\nBy OS:", dict(Counter(c["os"] for c in cmds)))
	print("By role:", dict(Counter(c["role"] for c in cmds)))
	print("By cat:", dict(Counter(c["cat"] for c in cmds)))

	if __name__ == "__main__":
	main()