review-ansible-playbooks

Ansible Playbook Review

Audit Ansible roles and playbooks for anti-patterns that ansible-lint does not catch — idempotency smells, handler misuse, unsafe secret handling, overly broad become: scope, and undisciplined tag use. This skill is strictly read-only, repo-local — it never SSHes to live hosts, never executes playbooks against real inventory, and never applies changes.

Invocation

The user runs /review-ansible-playbooks with an optional path to a single role (roles/common) or playbook (playbooks/site.yml). Without an argument, scan every role and playbook discoverable under the repository root.

Execution Steps

For each step, report PASS, FINDING, or SKIPPED (nothing in scope for that category).

1. Scope

Enumerate roles and playbooks under the argument path. Default globs when no argument is given:

• roles/*/tasks/*.yml, roles/*/handlers/*.yml, roles/*/defaults/*.yml, roles/*/vars/*.yml, roles/*/meta/main.yml, roles/*/templates/**/*.j2
• playbooks/**/*.yml, site.yml, *.ansible.yml
• Collection roles under collections/ansible_collections/*/*/roles/ when a requirements.yml or galaxy.yml is present

If no Ansible content is found, mark the run SKIPPED and stop.

2. Idempotency Smells

Grep task files for:

• ansible.builtin.command: or ansible.builtin.shell: (including short form command: / shell:) without any of creates:, removes:, changed_when:, or a comment justifying non-idempotency.
• ansible.builtin.lineinfile: applied to a file that is also managed by a nearby template: task — duplicate ownership.
• Repeated near-identical file: tasks that could collapse into a template: or copy: with a loop.
• raw: used outside bootstrap contexts (flag and require comment).
• check_mode: false used to hide a non-idempotent task — flag with the file:line and ask whether the logic can be rewritten idempotently.

3. Handler Misuse

• For each handler declared under roles/*/handlers/main.yml, grep the role's task files for matching notify: entries. Unused handler = FINDING.
• For each notify: entry in tasks, verify a handler of that exact name exists in the same role or is imported. Missing handler = FINDING.
• Handlers that themselves contain multi-step logic (more than one task of substantive work, or block: with several inner steps) should be promoted to a role or a tasks file included with import_tasks:.

4. Secret Handling

Grep for:

• Variables with name-stems password, passwd, secret, token, api_key, apikey, private_key, passphrase declared in plain defaults/main.yml, vars/main.yml, group_vars/*.yml, or host_vars/*.yml without a $ANSIBLE_VAULT; header.
• lookup('env', '...') used in contexts where the surrounding task lacks no_log: true.
• hashi_vault, aws_ssm, or community.hashi_vault.vault_read_secret lookups in tasks without no_log: true.
• Files named vault.yml, secrets.yml, or matching *secret*.yml that are not encrypted (first-line check for $ANSIBLE_VAULT;) and that appear in git ls-files (tracked).

Each hit is a HIGH FINDING.

5. Become Scope

• become: yes / become: true set at the play level when many tasks within do not require elevation — suggest moving become: onto the specific tasks that need it.
• become_user: omitted on an escalated task when root is not required (e.g. managing a service account's config).
• become_method: su without justification (prefer sudo).
• Nested become: contradictions between play, block, and task.

6. Tag Discipline

• Build a set of all tags declared via tags: on plays, roles, blocks, and tasks.
• Build a set of all tags referenced in documentation comments and playbook headers (# Usage: ansible-playbook ... --tags foo).
• Orphan tags (declared but never referenced in docs or CI) = FINDING.
• Tasks with no tags: in a playbook that otherwise uses tags heavily = FINDING.
• If the playbook header comment does not document the --tags contract and the playbook uses tags, FINDING.

7. Run ansible-lint (If Available)

Invoke ansible-lint --nocolor against the scoped path if the binary is installed. Include its output in a dedicated subsection. Do not replicate its rule checks in this skill's own findings — attribute each ansible-lint hit to ansible-lint. If the binary is missing, mark the subsection SKIPPED.

Output Format

## Ansible Playbook Review — <path or repo root>

### Scope
Roles scanned: N
Playbooks scanned: N

### Findings by Role / Playbook

#### roles/common
- [FINDING] tasks/bootstrap.yml:14 — shell: without creates / changed_when
- [FINDING] handlers/main.yml — handler "restart app" declared, never notified
- [HIGH] vars/main.yml:3 — plain-text "api_token" (should be vaulted)

#### roles/nginx
- [FINDING] tasks/main.yml — play-level become: yes but 6 of 11 tasks do not require root
- [FINDING] tasks/main.yml — tasks missing tags while other tasks in the role are tagged

#### playbooks/site.yml
- [FINDING] header comment does not document --tags contract; tags "deploy", "config" referenced by tasks

### ansible-lint Output
<captured output, or SKIPPED: ansible-lint not installed>

**Totals:** FINDING: N, HIGH: N
**Verdict:** <PASS | CONCERN | FAIL>

Verdict

• PASS: zero FINDINGs across all categories.
• CONCERN: one or more FINDINGs, no HIGH severity.
• FAIL: one or more HIGH findings (plain-text secret in tracked file, secret lookup without no_log:, unencrypted vault.yml / secrets.yml tracked in git).
• SKIPPED: no Ansible content in scope.

Rules

1. Read-only, repo-local. Never SSH. Never run ansible-playbook without --check. Never run ansible ad-hoc commands. Allowed commands: ansible-inventory --list, ansible-lint, ansible-playbook --check --diff. Nothing else.
2. No mutation. Do not propose ansible-playbook ... without --check. Do not propose ssh, rsync, or scp.
3. No fabrication. Do not invent role names, host names, tag names, or task contents. If scope is empty, mark SKIPPED rather than guessing.
4. Attribute ansible-lint findings to ansible-lint. Keep this skill's own output focused on the higher-level smells the linter misses.
5. Graceful SKIPPED. Missing ansible-lint, missing inventory, or empty scope each degrade to SKIPPED for the relevant step.
6. Report, do not fix. Surface findings; never auto-edit roles or playbooks.

$ARGUMENTS