generate-playbook
Write an Ansible playbook, inventory, and group_vars for a repeatable change across many network devices, then validate it locally before handing it back. Use when the user asks for Ansible by name, or asks how to push the same change to a fleet — VLANs, banners, NTP/AAA, SNMPv3, config backup, compliance checks — using cisco.ios, cisco.nxos, arista.eos, juniper.device (Junos), paloaltonetworks.panos, or fortinet.fortios. For Python (Nornir/Netmiko/Scrapli) use generate-automation, for Terraform use generate-terraform, for pyATS/pytest checks use generate-tests, for a CI pipeline use generate-pipeline, and for a single device's config use generate-config.
- License 许可
- MIT
- 最近更新
- 2026-11-02
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- generate-playbook
- description
- Write an Ansible playbook, inventory, and group_vars for a repeatable change across many network devices, then validate it locally before handing it back. Use when the user asks for Ansible by name, or asks how to push the same change to a fleet — VLANs, banners, NTP/AAA, SNMPv3, config backup, compliance checks — using cisco.ios, cisco.nxos, arista.eos, juniper.device (Junos), paloaltonetworks.panos, or fortinet.fortios. For Python (Nornir/Netmiko/Scrapli) use generate-automation, for Terraform use generate-terraform, for pyATS/pytest checks use generate-tests, for a CI pipeline use generate-pipeline, and for a single device's config use generate-config.
02 资源文件 1
- 参考文档references/ansible-collections.md4.6 KB
03 SKILL.md 正文
Ansible playbook generation
You write the playbook. Damira supplies the vendor facts and the gate: every playbook goes
through damira validate before you hand it back.
Step 1: Gather requirements
- Task — deploy VLANs, back up configs, push NTP/AAA/SNMPv3, compliance check
- Targets — how many devices, what platform and OS version
- Variables — VLAN IDs, interface names, server IPs
- Connection —
network_cli(SSH),httpapi(eAPI/NX-API), ornetconf
If the user named a different framework, use the sibling skill instead (see the description).
Step 2: Look up the modules
Run ~/.damira/bin/damira search-vendor-docs "<query>" --vendor <vendor> for the collection and module you plan to use, e.g.
"ansible cisco.ios ios_ntp_global parameters" with the vendor set. Module names and
parameters change between collection versions; this lookup is what keeps the playbook
from using a parameter that no longer exists. If the lookup fails, say so rather than
emitting parameters you could not verify.
Read references/ansible-collections.md (beside this file) for the collection per platform,
resource modules vs *_config, connection plugins, and check-mode caveats.
If the change is a standard baseline (NTP, AAA/TACACS+, SNMPv3, syslog) on IOS,
NX-OS, EOS or Junos, the plugin ships golden templates. Render the device lines with
--secrets ansible, so secret values (TACACS key, SNMP passphrases) come out as
{{ lookup('env', 'NAME') }} and are resolved when the playbook runs — never render real
secrets into your context or into a playbook:
~/.damira/bin/damira render <task> --platform <p> --vars vars.yml --secrets ansible
and push them with the platform's *_config module (Junos: juniper.device.junos_config
with the set lines under lines: — it takes set/delete lines directly and has no format
option; drop the # comment lines, which it rejects), instead of hand-writing the lines. Banners are the exception: use the
banner modules (cisco.ios.ios_banner, cisco.nxos.nxos_banner, arista.eos.eos_banner,
juniper.device.junos_banner) with the text as a variable — a
delimited banner login ^ ... ^ block does not fit *_config lines:.
Step 3: Write the playbook
- Inventory — YAML, device groups,
ansible_network_os+ansible_connectionper group - Credentials — Ansible Vault or env lookups (
lookup('env', 'NET_PASSWORD')), never inline - Tasks — resource modules (
state: merged) where one exists;*_configonly for lines with no resource module.gather_facts: false. Fully-qualified module names. - Error handling —
block/rescuearound anything that changes device state - Idempotency — safe to run repeatedly. Resource modules are;
*_configlines:only when they match the running config exactly. Golden AAA/SNMPv3 lines carrying secrets never do (IOS showskey 7, hidessnmp-server user), so they reportchangedevery run. Say so in the hand-back, or split them into a task withwhen:on a first-run/rotation flag
Save under playbooks/: playbooks/{task}.yml, playbooks/inventory/{group}.yml,
playbooks/group_vars/{group}.yml.
Validate loop
Delegate this loop to the damira-fixer subagent (pinned to a cheap model). Give it the path and --skill generate-playbook. It runs the command below, fixes only what the findings name, and stops after 3 rounds. Read its report: a change to a device command, module or resource is your call, not its. If the subagent isn't available, run the loop yourself as below.
The validator is ~/.damira/bin/damira validate. It runs locally (yamllint, ansible-playbook --syntax-check, ansible-lint --profile production) and makes no API call. --skill tags
the local workflow event with this skill.
~/.damira/bin/damira validate playbooks/ --skill generate-playbook --host cursor
- Run it on everything you wrote.
- Fix every FAIL and re-run — at most 3 rounds.
- Still failing after 3 rounds: stop, tell the user which checks fail and why. Do not hand it back as finished.
- A
skippedcheck means the tool is not installed — pass the install hint on; it is not a pass. AnUNVERIFIEDresult is not a pass either. couldn't resolve module/action '<ns>.<coll>.<module>'orsyntax-check[unknown-module]means the collection is not installed, not that the module is wrong. Do not rename or swap modules to make it go away: treat it like a skipped check, tell the user to runansible-galaxy collection install <ns>.<coll>(or-r collections/requirements.yml) and re-validate, and don't count it against the 3 rounds.
Step 4: Hand back
Lead with the dry run, then a single-device first run:
ansible-playbook -i playbooks/inventory/{group}.yml playbooks/{task}.yml --check --diff
# review the diff, then one device first:
ansible-playbook -i playbooks/inventory/{group}.yml playbooks/{task}.yml --limit <one-host>
State what validate reported (pass / warnings / skipped tools) and any assumption you made.
04 同仓库更多 Skills
upgrade-plan
Ask Damira to assess a version-to-version upgrade of a network platform and produce the MOP and change control. Use when the user names a product and a target release — "IOS-XE 17.6.5 to 17.9.5", "CUCM 12.5 SU7 to 15", "PAN-OS 10.2 to 11.1" — or asks what breaks, what the risk is, or how long a maintenance window needs to be. Covers CUCM, IOS-XE, IOS-XR, NX-OS, PAN-OS, FortiOS, Junos, ISE, ASA, WLC/Catalyst 9800. Not for platform-to-platform migration (CUCM to Webex Calling, on-prem to cloud) and not for diagnosing a fault after an upgrade — use troubleshoot for that.
troubleshoot
Ask Damira to diagnose an active, in-progress network fault. Use when the user reports something broken right now — an outage, a flapping interface, a BGP or OSPF adjacency down, dropped or blackholed traffic, failed calls, users unable to connect — or pastes show/log output for analysis. Also fires on direct questions like "why is BGP down" or "why is OSPF stuck in EXSTART". Returns vendor-specific diagnosis with exact CLI commands. Do NOT use for "how does X work" protocol questions, for reviewing a config that is not currently broken (use config-audit), or for planning an upgrade (use upgrade-plan).
source-of-truth-change
Turn intended state in the engineer's source of truth (NetBox or Nautobot) into a validated change, then hand it to the automation they already run — an Ansible Automation Platform job in check mode, a pyATS/Genie before-and-after diff, or a Terraform plan. Use when the user asks to make a change "from NetBox" or "from Nautobot", to push what the source of truth says, to bring devices in line with intended state, or to run a validated playbook through AAP or pyATS. Damira never applies the change; the engineer does. For a playbook with no source of truth use generate-playbook, for Terraform alone use generate-terraform, and for an active outage use troubleshoot.
lab-test-change
Test a network change in a throwaway ContainerLab lab before it goes near production — deploy the lab, apply the playbook or script, run show-command or pyATS/Genie checks, report pass/fail, and always destroy the lab. Use when the user says "test this in a lab first", "prove this change before the window", "dry-run this playbook against a lab", or wants evidence a change works before a CAB. Needs docker and containerlab on the engineer's machine; without them it validates only and says so. To only build a lab, use generate-lab; to write the change, use generate-playbook, generate-automation or generate-terraform.
generate-workbook
Build an Excel workbook (.xlsx) for a network procedure — a migration, platform migration, security audit, change control, or incident-response workbook — grounded in Damira evidence, with an HTML preview. Use when the user asks for a spreadsheet, workbook, tracker, or Excel version of a migration plan, cutover, audit findings, change record, or incident timeline. For a version-to-version upgrade, use upgrade-plan (it offers the workbook itself). Not for diagnosing an active fault — use troubleshoot.
generate-tests
Write network tests — pyATS/Genie learn, change, learn, diff jobs that prove a change did what it should and nothing else, and pytest tests for automation scripts — then validate them locally. Use when the user asks for pyATS, Genie, AEtest, a testbed file, pre/post change verification, state snapshots, or unit tests for a Nornir/Netmiko script. For the change itself use generate-playbook, generate-automation or generate-terraform.