JNPRAutomate/fw-skills-share / skills/firewall-config-diff
firewall-config-diff
VerifiedCompare two parsed firewall configurations by semantic intent rather than text. Use when checking drift, HA parity, pre/post-change results, migration fidelity, or round-trip conversion. Parse raw configs first; use a text diff for literal line changes.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 11.2 KB |
| agents/openai.yaml | config | 224 B |
| references/equivalence-rules.md | markdown | 18.1 KB |
| references/example-diff.md | markdown | 15.6 KB |
| references/runtime-intake.md | markdown | 5.4 KB |
references/example-diff.md
View on GitHub# Worked Examples: Firewall Config Diff
## Contents
- [Scenario 1 — Same-vendor drift (Cisco ASA → Cisco ASA)](#scenario-1--same-vendor-drift-cisco-asa--cisco-asa)
- [Scenario 2 — Cross-vendor parity (Cisco ASA-shaped vs Juniper SRX-shaped)](#scenario-2--cross-vendor-parity-cisco-asa-shaped-vs-juniper-srx-shaped)
- [Scenario 3 — Round-trip conversion self-test (Cisco schema A vs re-parsed SRX schema B)](#scenario-3--round-trip-conversion-self-test-cisco-schema-a-vs-re-parsed-srx-schema-b)
Three worked diffs produced by the `firewall-config-diff` skill, each emitting the exact
**Output & Verdict** template from `SKILL.md` and applying the pairing/classification rules
in `references/equivalence-rules.md`. Every comparison is **semantic** (by meaning, not text):
items pair by value / member-set / match-and-action tuple, never by name or position.
The diff markers are fixed:
- `+ [B]` — added (present only in B)
- `- [A]` — removed (present only in A)
- `~ [A→B]` — changed (same semantic identity, ≥1 differing attribute)
Each scenario ends with a single `Parity verdict:` line. Secrets (PSK / certs / passwords)
are never printed — only a presence/changed flag.
- **Scenario 1** — same-vendor **drift** (Cisco ASA baseline vs a modified copy).
- **Scenario 2** — cross-vendor **parity** (ASA-shaped vs SRX-shaped, semantically equal on
comparable sections; VPN + threat profiles `not-comparable`).
- **Scenario 3** — **round-trip** conversion self-test (Cisco schema A vs the schema from
re-parsing the SRX that the `firewall-config-conversion` skill emits for that fixture).
---
## Scenario 1 — Same-vendor drift (Cisco ASA → Cisco ASA)
**Source for A:** the real parsed Cisco fixture (`parsing-cisco-configs`
`references/fixture-expected-output.json`, `metadata.source_vendor: cisco_asa`). It contains exactly:
zones `outside` / `inside`; interfaces `GigabitEthernet0/0` (203.0.113.2/24) /
`GigabitEthernet0/1` (10.0.0.1/24); address objects `WEB` (10.0.1.10/32) and `INSIDE-NET`
(10.0.0.0/16); service object `HTTPS-ALT` (tcp/8443); security policy `OUTSIDE-IN-1`
(any → WEB, tcp/443, allow, log at session-end); source NAT `INSIDE-NET-auto-nat`
(inside → outside, INSIDE-NET behind egress interface); one default static route.
**B is that exact fixture with three clearly-labeled edits — nothing else changes:**
1. **Add one permit policy** `INSIDE-WEB-1`: inside → outside, src `INSIDE-NET`, dst `any`,
service `HTTPS-ALT` (tcp/8443, the previously-unreferenced fixture service object), allow.
2. **Change one address object value:** `WEB` 10.0.1.10/32 → 10.0.1.20/32.
3. **Toggle logging on `OUTSIDE-IN-1`:** `log_end` true → false.
This is a drift framing (A = baseline, B = current). Because both sides are the same vendor
and the object names are stable, pairs are anchored by their stable identity and an attribute
delta is reported as `~ changed` (vs cross-vendor, where addresses pair strictly by value).
```text
Section: address_objects unchanged 1 added 0 removed 0 changed 1
~ [A→B] WEB: value 10.0.1.10/32 → 10.0.1.20/32
(INSIDE-NET 10.0.0.0/16 unchanged; the WEB value shift propagates to every policy that
references WEB — see OUTSIDE-IN-1 effective destination below)
Section: service_objects unchanged 1 added 0 removed 0 changed 0
(HTTPS-ALT tcp/8443 unchanged — it is now referenced by the new policy, but the object
definition itself did not change)
Section: security_policies unchanged 1 added 1 removed 0 changed 1 | order: 0 reordered
(the fixture's implicit default-deny pairs implicit-to-implicit on both sides — unchanged)
+ [B] INSIDE-WEB-1: inside → outside, INSIDE-NET → any, tcp/8443 (HTTPS-ALT), allow
~ [A→B] OUTSIDE-IN-1 (any → WEB, tcp/443, allow): log_end true → false
(effective destination also moved with WEB: 10.0.1.10/32 → 10.0.1.20/32)
Section: nat_rules unchanged 1 added 0 removed 0 changed 0
Section: zones unchanged 2 added 0 removed 0 changed 0
Section: interfaces unchanged 2 added 0 removed 0 changed 0
Section: static_routes unchanged 1 added 0 removed 0 changed 0
Not comparable: none (same vendor — all sections isomorphic)
Parity verdict: DIFFERENCES FOUND (3) (A=cisco_asa, B=cisco_asa)
```
The verdict count is 3: one added policy, one changed address object, one changed policy
(logging). Rule order did not change (the new policy appended), so no reorder is reported.
---
## Scenario 2 — Cross-vendor parity (Cisco ASA-shaped vs Juniper SRX-shaped)
Two **different-vendor** sides describing the same intent. Per `equivalence-rules.md`, the
constructs the vendors model differently are normalized to a common form *before* pairing,
and features with no equivalent on the other side are flagged `not-comparable` — never a
false diff.
**A — ASA-shaped snippet:**
```text
interface GigabitEthernet0/0
nameif outside
security-level 0
object network web-srv
host 10.0.1.10
access-list OUTSIDE_in extended permit tcp any object web-srv eq https
access-group OUTSIDE_in in interface outside
crypto ikev2 policy 1
encryption aes-256
integrity sha256
group 14
lifetime seconds 86400
crypto map OUT_MAP 10 match address VPN_TRAFFIC
crypto map OUT_MAP 10 set peer 198.51.100.7
crypto map OUT_MAP 10 set ikev2 ipsec-proposal AES256-SHA256
! pre-shared-key <redacted> (secret never compared by value)
! classic ASA has no NGFW / UTM profile object
```
**B — SRX-shaped snippet:**
```text
set security address-book global address host-A 10.0.1.10/32
set security policies global policy ALLOW-WEB match from-zone outside
set security policies global policy ALLOW-WEB match source-address any
set security policies global policy ALLOW-WEB match destination-address host-A
set security policies global policy ALLOW-WEB match application junos-https
set security policies global policy ALLOW-WEB then permit
set security policies global policy ALLOW-WEB then permit application-services security-intelligence-policy SECINTEL
set security ike proposal IKE-PROP encryption-algorithm aes-256-cbc
set security ike proposal IKE-PROP authentication-algorithm sha-256
set security ike proposal IKE-PROP dh-group group14
set security ike proposal IKE-PROP lifetime-seconds 86400
set security ike policy IKE-POL proposals IKE-PROP
set security ike gateway GW-SITEB ike-policy IKE-POL
set security ike gateway GW-SITEB address 198.51.100.7
set security ike gateway GW-SITEB version v2-only
set security ipsec vpn VPN-SITEB bind-interface st0.0
set security ipsec vpn VPN-SITEB ike gateway GW-SITEB
# pre-shared-key <redacted> (secret never compared by value)
```
**Normalization applied before pairing:**
- `web-srv host 10.0.1.10` (A) and `host-A 10.0.1.10/32` (B) carry the **same value**
10.0.1.10/32 — different names, equal object.
- ASA `eq https` (tcp/443) and SRX `junos-https` both normalize to canonical app `https` —
equal, not a diff.
- ASA derives a pseudo-zone `outside` from `nameif outside` (`security-level 0`); the SRX side
uses a `security policies global` rule scoped with `match from-zone outside`, so the ingress
scope pairs — the ASA ACL applied inbound on `outside` and the zone-scoped global rule cover
the same traffic. (An *unscoped* global rule would be broader than the interface-scoped ACL
and would pair as `changed`, not equal.) The ASA `security-level` trust ordering and
its implicit high→low permit have no SRX equivalent and are `not-comparable`.
- VPN crypto **proposal fields** normalize and compare: ASA `crypto ikev2 policy` (aes-256 /
sha256 / DH group14 / lifetime 86400) and SRX `IKE-PROP` (aes-256-cbc / sha-256 / group14 /
86400) are **equal**. Only the tunnel **model** differs (see vpn_tunnels below).
```text
Section: address_objects unchanged 1 added 0 removed 0 changed 0
(10.0.1.10/32 present on both — paired by value, not name)
Section: security_policies unchanged 1 added 0 removed 0 changed 0
(any → 10.0.1.10/32, https, allow — identical tuple after app + zone normalization)
Section: vpn_tunnels unchanged 1 added 0 removed 0 changed 0 (proposal fields only)
(peer 198.51.100.7 pairs; IKE proposal normalizes EQUAL both sides — enc aes-256,
auth sha-256, DH group14, lifetime 86400; psk present on both, compared by presence only.
The tunnel MODEL — ASA policy-based crypto-map vs SRX route-based st0 / proxy-id vs
traffic-selector — is `not-comparable` and listed below, not a diff.)
Section: security_services not-comparable
(threat-inspection attachment is a coarse present/absent boolean only between NGFW-capable
vendors; classic ASA has no UTM/NGFW object, so to/from ASA the attachment and all
profile contents are `not-comparable` — SRX `permit application-services` SecIntel/IDP
has nothing to pair against on A. Not reported as a missing policy.)
Not comparable: vpn_tunnels tunnel-model (ASA crypto-map vs SRX route-based st0 / proxy-id vs
traffic-selector — proposal fields above DID compare equal), security_services (no UTM on
classic ASA — threat attachment not-comparable to/from ASA), physical interface names,
ASA `security-level` trust ordering and implicit high→low permit
Parity verdict: EQUIVALENT (A=cisco_asa, B=srx)
```
The comparable sections — including the VPN **proposal fields** — are equivalent; only the
non-isomorphic pieces (the VPN tunnel-model, and threat profiles to/from a classic ASA that
has no UTM) are excluded from add/removed/changed rather than reported as false differences.
---
## Scenario 3 — Round-trip conversion self-test (Cisco schema A vs re-parsed SRX schema B)
This measures **conversion fidelity**. A is the original Cisco fixture schema. B is produced
by taking the SRX `set` config that the `firewall-config-conversion` skill emits for that same
fixture, then re-parsing it with `parsing-srx-configs` back into the intermediate schema. If
the conversion were lossless, A and B would be `EQUIVALENT` on every comparable section. The
diff surfaces exactly where fidelity is lost. (Per `SKILL.md`, this example stays
self-contained — the conversion skill is named, never path-referenced.)
**A — original Cisco fixture (schema, abridged):**
```json
{
"address_objects": [
{"name": "WEB", "type": "host", "value": "10.0.1.10/32"},
{"name": "INSIDE-NET", "type": "subnet", "value": "10.0.0.0/16"}
],
"service_objects": [
{"name": "HTTPS-ALT", "protocol": "tcp", "port_range": "8443"}
],
"security_policies": [
{"name": "OUTSIDE-IN-1", "src_zones": ["outside"], "dst_zones": ["any"],
"src_addresses": ["any"], "dst_addresses": ["WEB"], "services": ["tcp/443"],
"action": "allow", "log_end": true}
],
"interfaces": [
{"name": "GigabitEthernet0/0", "zone": "outside", "ip": "203.0.113.2/24"},
{"name": "GigabitEthernet0/1", "zone": "inside", "ip": "10.0.0.1/24"}
]
}
```
**B — schema from re-parsing the emitted SRX (abridged):** the conversion emits
`set security address-book global address WEB 10.0.1.10/32`, the policy as
`junos-https` with `then log session-close`, `HTTPS-ALT` as a **custom application**
(not a service object — it re-parses into `applications[]` with `vendor_name: HTTPS-ALT` and
no canonical key, `confidence: 0.0`), and interfaces remapped to `ge-0/0/0` / `ge-0/0/1`.
Re-parsed:
```json
{
"address_objects": [
{"name": "WEB", "type": "host", "value": "10.0.1.10/32"},
{"name": "INSIDE-NET", "type": "subnet", "value": "10.0.0.0/16"}
],
"service_objects": [],
"applications": [
{"vendor_name": "HTTPS-ALT", "canonical": null, "confidence": 0.0, "category": "custom"}
],
"security_policies": [
{"name": "010-OUTSIDE-IN-1", "src_zones": ["outside"], "dst_zones": ["any"],
"src_addresses": ["any"], "dst_addresses": ["WEB"], "applications": ["https"],
"action": "permit", "log_end": true}
],
"interfaces": [
{"name": "ge-0/0/0.0", "zone": "outside", "ip": "203.0.113.2/24"},
{"name": "ge-0/0/1.0", "zone": "inside", "ip": "10.0.0.1/24"}
]
}
```
**Pairing notes (semantic):** addresses pair by value (both 10.0.1.10/32, 10.0.0.0/16);
the policy tuple matches after normalization — `junos-https` == canonical `https` == ASA
tcp/443, `permit` == `allow`, `log session-close` == `log_end: true`, and the rename
`OUTSIDE-IN-1` → `010-OUTSIDE-IN-1` is **not** a diff (name-insensitive); interfaces pair by
**address**, since physical names are `not-comparable`.
```text
Section: address_objects unchanged 2 added 0 removed 0 changed 0
(WEB 10.0.1.10/32, INSIDE-NET 10.0.0.0/16 — clean round-trip by value)
Section: security_policies unchanged 1 added 0 removed 0 changed 0 | order: 0 reordered
(tuple any → WEB/https/allow/log preserved; junos-https↔tcp/443, permit↔allow,
log session-close↔log_end all normalize equal; numeric-prefix rename is not a diff)
Section: nat_rules unchanged 1 added 0 removed 0 changed 0
Section: static_routes unchanged 1 added 0 removed 0 changed 0
Section: zones unchanged 2 added 0 removed 0 changed 0
(zone names outside/inside preserved by the conversion — isomorphic)
Section: service_objects unchanged 0 added 0 removed 1 changed 0
- [A] HTTPS-ALT tcp/8443: ASA `object service` re-emitted by the conversion as an SRX
custom `application`, so the service_objects section genuinely loses it — a real
section-shape fidelity gap, counted here as the one comparable difference. Its intent
is recovered by value (tcp/8443) under B's `applications`, but that B-side custom app
has no canonical key (`confidence: 0.0`) and so is `not-comparable`, not a B add.
Section: interfaces unchanged 2 added 0 removed 0 changed 0 (paired by address)
(both addresses 203.0.113.2/24 and 10.0.0.1/24 pair cleanly; the only delta is the
physical name remap GigabitEthernet0/x → ge-0/0/x.0, which is `not-comparable` across
platforms — listed below, NOT counted as a changed pair)
Not comparable: physical interface names (GigabitEthernet0/x vs ge-0/0/x.0 — paired by IP, the
name remap is excluded from the diff count); B-side custom application HTTPS-ALT
(unresolved, no Cisco canonical equivalent); ASA `nameif` + `security-level` semantics
(implicit high→low permit has no SRX equivalent — the conversion drops it, so SRX
default-deny would block inside→outside until an explicit permit policy is added; this
intent loss is a fidelity gap, not a schema add/remove)
Parity verdict: DIFFERENCES FOUND (1) (A=cisco_asa, B=srx)
```
The verdict count is 1: the single comparable difference is the `service_objects` removal
(removed 1). The interface name remap and the B-side custom application are `not-comparable`
and are listed only on the `Not comparable:` line — excluded from added/removed/changed and
from the verdict number.
**Fidelity read-out:** the security-relevant content round-trips cleanly — addresses,
the policy match-and-action tuple, NAT, zones, routes, and the interface addresses are all
`EQUIVALENT`. The gaps are **representational, not behavioral mismatches in the comparable
rule set**: (1) the ASA `object service` migrates into SRX's custom-application section, the
one counted difference (recovered by value, but the section shape is not preserved);
(2) interface names are platform-bound and remap — `not-comparable`, not a diff; (3) the ASA
`security-level` implicit-permit semantics have no SRX equivalent and are dropped by the
conversion — the single biggest fidelity risk, since it means an explicit inside→outside
permit must be added on SRX. None of (2) or (3) is a false diff: each is recovered by value or
flagged `not-comparable`.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| slack-gif-creatoranthropics/skillsKnowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack." | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| ponytailDietrichGebert/ponytailLazy senior dev mode for any coding task: the smallest change that fully solves it, and a reply a busy human understands in one read. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailHarvest every shortcut: comment into one debt ledger, so deferrals get tracked instead of forgotten. One-shot report. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-debtDietrichGebert/ponytailList every `shortcut:` comment (and older `ponytail:` ones) as a debt ledger. One-shot report, changes nothing. Use for "ponytail debt", "what did ponytail defer", "list the shortcuts", /ponytail-debt. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |