diff --git a/docs/superpowers/plans/2026-07-28-ansible-reconciliation.md b/docs/superpowers/plans/2026-07-28-ansible-reconciliation.md new file mode 100644 index 0000000..607051b --- /dev/null +++ b/docs/superpowers/plans/2026-07-28-ansible-reconciliation.md @@ -0,0 +1,451 @@ +# Ansible-arch ↔ Machine Reconciliation Implementation Plan + +> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking. + +**Goal:** Bring the `ansible-arch` repo and the machine it manages back into alignment: commit the in-flight working-tree changes, fix the `become=True` global-default bug that caused root-owned dotfiles symlinks, and codify already-installed-but-undeclared packages into the right roles. + +**Architecture:** No new files or abstractions. This is a reconciliation pass over existing `roles/*/tasks/*.yml` files and `ansible.cfg`. Every privileged task in the repo already carries an explicit `become: true` — the audit confirmed the *only* structural bug is the global default in `ansible.cfg`, so that fix is a one-line change, not a per-task sweep. Verification is `ansible-playbook --syntax-check` / `--check --diff` where the modules support check mode, and one real `ansible-playbook site.yml` run at the end (safe: pacman tasks are no-ops for already-installed packages, dotfiles-role symlink recreation is idempotent by design). + +**Tech Stack:** Ansible, Arch Linux (pacman, AUR via `kewlfft.aur` + `paru`), GNU Stow. + +--- + +### Task 1: Commit the in-flight working-tree diff + +**Files:** +- Modify (already modified in working tree, no further edits needed): `dotfiles/hypr/.config/hypr/hyprland.conf`, `roles/ai/tasks/main.yaml`, `roles/common/defaults/main.yml`, `roles/common/tasks/main.yml`, `roles/devops/tasks/main.yml`, `roles/dotfiles/defaults/main.yml`, `roles/dotfiles/tasks/main.yml`, `roles/hyprland/tasks/main.yml`, `roles/multimedia/tasks/main.yml` +- Add (untracked): `dotfiles/hypr/.config/hypr/hyprlock.conf`, `dotfiles/hypr/.config/hypr/monitors.sh`, `dotfiles/kanshi/`, `dotfiles/starship/` + +This working-tree diff already reflects the intended current state (verified during brainstorming — it's the hyprland.conf tweaks, new dotfiles stow packages, and new role packages that led to this reconciliation). Nothing to write here, just commit it. + +- [ ] **Step 1: Review the diff one more time** + +```bash +git status +git diff --stat +``` + +Expected: the same 9 modified files and 4 untracked paths listed in the design spec's "Context" section. + +- [ ] **Step 2: Stage and commit** + +```bash +git add dotfiles/hypr/.config/hypr/hyprland.conf \ + dotfiles/hypr/.config/hypr/hyprlock.conf \ + dotfiles/hypr/.config/hypr/monitors.sh \ + dotfiles/kanshi/ dotfiles/starship/ \ + roles/ai/tasks/main.yaml \ + roles/common/defaults/main.yml \ + roles/common/tasks/main.yml \ + roles/devops/tasks/main.yml \ + roles/dotfiles/defaults/main.yml \ + roles/dotfiles/tasks/main.yml \ + roles/hyprland/tasks/main.yml \ + roles/multimedia/tasks/main.yml +git commit -m "feat: formalize in-flight hyprland/dotfiles/role changes" +git status +``` + +Expected: `git status` shows "nothing to commit, working tree clean" (modulo files touched by later tasks). + +--- + +### Task 2: Fix the `become=True` global default + +**Files:** +- Modify: `ansible.cfg:11-15` + +**Context:** `ansible.cfg` currently has: + +```ini +[privilege_escalation] +become = True +become_method = sudo +become_user = root +become_ask_pass = False +``` + +Every task across every role that actually needs root (`pacman:`, `systemd:`, `user:`, `getent:`, `chsh`) already has an explicit `become: true` set on it — confirmed by reading every `roles/*/tasks/*.yml` file. Every task that must run unprivileged (AUR installs via `kewlfft.aur.aur`, the `paru` makepkg bootstrap, `oh-my-zsh` install, dotfiles `stow`/`file` tasks, `blockinfile` edits to `.zshrc`/`.profile`) already correctly omits `become`. The global `become = True` default was silently overriding all of those correctly-unprivileged tasks to root — that's why `~/.config/hypr`, `~/.config/zellij`, and `~/.config/starship.toml` symlinks are owned by `root`. + +So the fix is exactly one line. + +- [ ] **Step 1: Remove the global become default** + +Edit `ansible.cfg`, changing: + +```ini +[privilege_escalation] +become = True +become_method = sudo +become_user = root +become_ask_pass = False +``` + +to: + +```ini +[privilege_escalation] +become_method = sudo +become_user = root +become_ask_pass = False +``` + +- [ ] **Step 2: Syntax-check the playbook** + +```bash +ansible-playbook site.yml --syntax-check +``` + +Expected: `playbook: site.yml` with no errors. + +- [ ] **Step 3: Dry-run the parts that support check mode** + +```bash +ansible-playbook site.yml --check --diff +``` + +Expected: pacman/systemd/user tasks report no changes (packages already installed, services already enabled). The `stow`/`command` tasks in the dotfiles role will show as `skipped` (partial check-mode support) rather than simulate the symlink recreation — that's expected and covered by the real run in Task 6. + +- [ ] **Step 4: Commit** + +```bash +git add ansible.cfg +git commit -m "fix: stop forcing become=True globally, rely on per-task become" +``` + +--- + +### Task 3: Add missing packages to the `ctf` role + +**Files:** +- Modify: `roles/ctf/tasks/main.yml` + +Current content: + +```yaml +# tasks file for ctf +- name: Install CTF tools (official repos) + become: true + pacman: + name: + - wireshark-qt + - nmap + - openbsd-netcat + - binwalk + - gobuster + state: present +``` + +`gitleaks`, `picocom`, and `python-pyserial` are official-repo packages. `nuclei-bin` is AUR-only. + +- [ ] **Step 1: Add the official-repo packages** + +Edit `roles/ctf/tasks/main.yml` to: + +```yaml +# tasks file for ctf +- name: Install CTF tools (official repos) + become: true + pacman: + name: + - wireshark-qt + - nmap + - openbsd-netcat + - binwalk + - gobuster + - gitleaks + - picocom + - python-pyserial + state: present + +- name: Install CTF AUR packages + kewlfft.aur.aur: + name: "{{ item }}" + use: "{{ aur_helper | default('paru') }}" + state: present + loop: + - nuclei-bin +``` + +- [ ] **Step 2: Syntax-check** + +```bash +ansible-playbook site.yml --syntax-check +``` + +Expected: no errors. + +- [ ] **Step 3: Dry-run just this role** + +```bash +ansible-playbook site.yml --check --diff --tags ctf +``` + +Expected: `pacman` task reports `ok` (all packages already installed, since the audit confirmed they're present via `pacman -Qeq`). The AUR task will show as `skipped` in check mode (module doesn't support check mode) — that's fine, it's covered by the Task 6 real run. + +- [ ] **Step 4: Commit** + +```bash +git add roles/ctf/tasks/main.yml +git commit -m "feat(ctf): add gitleaks, nuclei-bin, picocom, python-pyserial" +``` + +--- + +### Task 4: Add missing packages to the `common` role + +**Files:** +- Modify: `roles/common/tasks/main.yml:19-53` + +Current "Install base development tools" task (lines 19-53): + +```yaml +- name: Install base development tools + become: true + pacman: + name: + - base-devel + - git + - curl + - wget + - rsync + - unzip + - zip + - less + - man-db + - texinfo + - which + - zsh + - stow + - ansible + - zoxide + - atuin + - broot + - bottom + - hyperfine + - tokei + - git-delta + - yazi + - ueberzugpp + - tealdeer + - ledger + - btop + - htop + - bind + - socat + - iperf3 + state: present +``` + +Add `nano`, `vim`, `sshpass`, `smartmontools` here. `cargo-audit` isn't a pacman package (it's installed via `cargo install`), so it goes in the separate "Install Rust toolchain and Python uv" task as a follow-up `cargo` step. + +- [ ] **Step 1: Add nano, vim, sshpass, smartmontools** + +Edit the `name:` list in "Install base development tools" (`roles/common/tasks/main.yml:22-51`), adding after `iperf3`: + +```yaml + - iperf3 + - nano + - vim + - sshpass + - smartmontools + state: present +``` + +- [ ] **Step 2: Add cargo-audit via cargo install** + +Current "Install Rust toolchain and Python uv" task (`roles/common/tasks/main.yml:55-68`): + +```yaml +- name: Install Rust toolchain and Python uv + become: true + pacman: + name: + - rust + - eza + - bat + - ripgrep + - fd + - starship + - cargo + - cargo-binstall + - uv + state: present +``` + +Add a new task immediately after it: + +```yaml +- name: Install cargo-audit + command: cargo install cargo-audit --locked + args: + creates: "{{ archsetup_user_home }}/.cargo/bin/cargo-audit" +``` + +No `become` — `cargo install` must run as the invoking user, installing into `~/.cargo/bin`. + +- [ ] **Step 3: Syntax-check** + +```bash +ansible-playbook site.yml --syntax-check +``` + +Expected: no errors. + +- [ ] **Step 4: Dry-run** + +```bash +ansible-playbook site.yml --check --diff --tags common +``` + +Expected: pacman task `ok`, no changes (all four packages already installed per the audit). The `cargo-audit` command task will show as `skipped` in check mode — verified for real in Task 6 (`cargo-audit` is already on the machine, per the audit, so it'll be a no-op there too via the `creates` guard). + +- [ ] **Step 5: Commit** + +```bash +git add roles/common/tasks/main.yml +git commit -m "feat(common): add nano, vim, sshpass, smartmontools, cargo-audit" +``` + +--- + +### Task 5: Remove duplicate `docker` package from `cs_student` role + +**Files:** +- Modify: `roles/cs_student/tasks/main.yml:1-17` + +`devops` role (`roles/devops/tasks/main.yml:6`) already declares `docker`. Current `cs_student` task: + +```yaml +# tasks file for cs_student +- name: Install CS student tools + become: true + pacman: + name: + - neovim + - emacs + - python + - nodejs + - npm + - texlive-bin + - texlive-basic + - jupyter-nbconvert + - docker + - code + - github-cli + state: present +``` + +- [ ] **Step 1: Remove the `docker` line** + +Edit to: + +```yaml +# tasks file for cs_student +- name: Install CS student tools + become: true + pacman: + name: + - neovim + - emacs + - python + - nodejs + - npm + - texlive-bin + - texlive-basic + - jupyter-nbconvert + - code + - github-cli + state: present +``` + +- [ ] **Step 2: Syntax-check** + +```bash +ansible-playbook site.yml --syntax-check +``` + +Expected: no errors. + +- [ ] **Step 3: Commit** + +```bash +git add roles/cs_student/tasks/main.yml +git commit -m "refactor(cs_student): drop duplicate docker package, already in devops role" +``` + +--- + +### Task 6: Remove the dead duplicate handler file + +**Files:** +- Delete: `roles/common/handlers/main.yaml` + +**Context:** `roles/common/handlers/` contains both `main.yml` and `main.yaml`. Ansible's role loader (`ansible/playbook/role/__init__.py:_load_role_yaml`, confirmed by reading the installed Ansible source) searches extensions in the fixed order `['.yml', '.yaml', '.json']` and stops at the first match for handlers (`allow_dir=False`). That means `main.yml` — which already has `become: true` and is the one referenced correctly — is the only file Ansible ever loads. `main.yaml` is silently dead: it's never parsed, and its font-cache handler task (which lacks `become: true` and would break under the Task 2 fix if it were ever loaded) is orphaned. + +Since `main.yml` is confirmed live and already correct, `main.yaml` is pure dead weight causing confusion about which handler is authoritative — remove it. + +- [ ] **Step 1: Confirm main.yml is the active file** + +```bash +cat roles/common/handlers/main.yml +``` + +Expected output: + +```yaml +- name: Update font cache + become: true + command: fc-cache -fv +``` + +- [ ] **Step 2: Delete the dead duplicate** + +```bash +git rm roles/common/handlers/main.yaml +``` + +- [ ] **Step 3: Syntax-check** + +```bash +ansible-playbook site.yml --syntax-check +``` + +Expected: no errors (confirms `main.yml` alone still satisfies the `notify: Update font cache` handler reference in `roles/common/tasks/main.yml`). + +- [ ] **Step 4: Commit** + +```bash +git commit -m "chore(common): remove dead duplicate handlers/main.yaml (main.yml is the one Ansible loads)" +``` + +--- + +### Task 7: Real playbook run and final verification + +This is the only step in the plan that touches live machine state beyond `--check` dry-runs. It's expected to be safe and mostly a no-op: every package involved is already installed (confirmed via `pacman -Qeq` during the audit), and the dotfiles role's "Remove existing config files before stowing" task unconditionally removes and recreates its symlink targets on every run by design — that's what fixes the root-ownership bug, with no manual `rm`/`chown` needed. + +- [ ] **Step 1: Run the full playbook** + +```bash +ansible-playbook site.yml +``` + +Expected: playbook completes with `failed=0`. Some tasks report `changed` (dotfiles symlinks recreated, any packages that were pulled in by Tasks 3-4 but not yet actually reconciled by pacman state). + +- [ ] **Step 2: Verify symlink ownership** + +```bash +ls -la ~/.config/hypr ~/.config/zellij ~/.config/starship.toml ~/.config/kanshi +``` + +Expected: all four owned by `tudattr`, not `root`. + +- [ ] **Step 3: Verify no drift remains** + +```bash +ansible-playbook site.yml --check --diff +git status +``` + +Expected: second `--check` run reports no further changes (idempotent), `git status` shows a clean working tree. diff --git a/docs/superpowers/specs/2026-07-28-ansible-reconciliation-design.md b/docs/superpowers/specs/2026-07-28-ansible-reconciliation-design.md index 5042c46..d8e3721 100644 --- a/docs/superpowers/specs/2026-07-28-ansible-reconciliation-design.md +++ b/docs/superpowers/specs/2026-07-28-ansible-reconciliation-design.md @@ -105,6 +105,12 @@ single-value `name:` key the audit grep didn't match). **Minor cleanup:** remove the duplicate `docker` package entry from the `cs_student` role — `devops` already declares it. +**Dead file found during planning:** `roles/common/handlers/` contains both +`main.yml` and `main.yaml`. Ansible's role loader searches handler file +extensions in a fixed order and stops at the first match, so `main.yml` +(which already has `become: true`) is the only one ever loaded — `main.yaml` +is dead and orphaned. Delete it. + ## Verification - `ansible-playbook site.yml --check --diff` runs clean (no unexpected