angular-cop

Files (10)

• SKILL.md
• clean-architecture.md
• design-system.md
• enforcement-tooling.md
• enforcement.md
• flurryx.md
• output-format.md
• rxjs.md
• signals.md
• typescript-strict.md

SKILL.md (first lines)

# angular-cop

Pre-merge review. Compares HEAD vs `origin/<target>`. Angular-aware. Project-aware (reads `AGENTS.md`). Tooling-aware (runs lint + tsc).

## When to Activate

- `/cop-review <target>` slash command
- User asks for review before merging a PR
- User specifies a target branch and wants a diff review
- angular-cop agent is invoked

## Inputs

| Arg | Required | Default | Meaning |
|---|---|---|---|
| `<target>` | yes | — | Target branch (e.g. `main`, `develop`, `release/x`) |
| `--level` | no | auto | `junior` (verbose teaching) or `senior` (terse). Auto = senior. |
| `--scope` | no | all | Comma list: `signals,rxjs,arch,flurryx,ts,design,ddd,a11y` |
| `--no-tools` | no | false | Skip lint + tsc (static review only) |

## Hard Rules

1. **Read-only.** Never patch code. Output report only.
2. **Diff window:** `git merge-base HEAD origin/<target>`..`HEAD`. Never review changes already on target.
3. **Confidence ≥ 80%.** Skip uncertain findings. Use `❓` instead of a speculative `🔴 Important [bug]:`. ❓ never blocks.
4. **Project rules win.** `AGENTS.md` overrides this skill. Re-read on every run; do not cache between sessions.
5. **flurryx ground truth:** load the [[flurryx]] skill before flagging state-management code. Do not invent APIs.
6. **No fluff.** No "great work", no restating what the diff already shows.

## Pipeline

```
1. Parse args -> target, level, scope
2. git fetch <remote> <target>          (silent; --quiet)
3. base = git merge-base HEAD <remote>/<target>
4. changed = git diff --name-status base..HEAD
5. Load <repo>/AGENTS.md (if exists) -> project rules
6. For each changed file:
     - Skim full file (not just hunk) for context
     - Apply relevant sub-checklists by extension/role:
          *.component.ts / *.component.html / *.component.css -> signals.md, rxjs.md, clean-architecture.md, design-system.md, a11y
         *.facade.ts / *.store.ts -> flurryx.md, clean-architecture.md
         *.adapter.ts / *.port.ts -> clean-architecture.md
         *.ts                     -> typescript-strict.md
         domain/**/*.ts           -> angular-ddd
     - Apply skill `comment-judge` (REVIEW mode) to every added/changed comment
     - If --scope includes ddd && domain/ code changed: load [[angular-ddd]] (review-checklist.md) for tactical/strategic DDD checks
7. If !--no-tools:
     - npm run lint -- --quiet (or eslint --quiet) on changed files
     - npx tsc --noEmit (full project; abort early on first 50 errors)
8. Aggregate findings -> render via output-format.md
```

## Tags

| Badge | Tag | Category | Blocks |
|---|---|---|---|
| 🔴 | Important | [bug] broken behavior, runtime crash, data loss | yes |
| 🔴 | Important | [sec] security risk (XSS, leaked secret, auth bypass) | yes |
| 🔴 | Important | [arch] BLOCK architecture / layering rule | yes |
| 🔴 | Important | [comment] confirmed false/stale or forbidden comment; [format] failed format check | yes |
| 🟡 | Nit | [risk] works today, fragile tomorrow (leak, race, missing teardown) | no, unless the finding cites an AGENTS.md hard rule |
| 🟡 | Nit | [nit] style, naming, micro-optim | no |
| ⚪ | Pre-existing | any category already on the base | no |
| ❓ | question | [q] genuine question | never |

## Sub-pages (read on demand)

- [[angular-cop-enforcement]] — BLOCK vs WARN tag checklist (load always)
- [[angular-cop-enforcement-tooling]] — ESLint flat config + architecture plugins for app repos
- [[angular-cop-signals]] — Angular signals, change detection, OnPush, computed, no-method-in-template
- [[angular-cop-rxjs]] — RxJS hygiene, takeUntilDestroyed, async pipe, leak patterns
- [[angular-cop-clean-architecture]] — facade / use-case / port / adapter / store boundaries
- [[angular-cop-flurryx]] — flurryx-specific rules (decorator order, keyed stores, no manual Record updates)
- [[angular-cop-typescript-strict]] — no `any`, immutability, narrowing, no `!`, readonly
- [[angular-cop-design-system]] — Kourou v0 design-token / color / spacing / typography tags (load for *.html and template-bearing files)
- [[angular-ddd]] — DDD tactical patterns (entities, value objects, aggregates, repositories) and strategic design (bounded contexts, ubiquitous language, ACL) for domain-layer code
- [[angular-cop-output-format]] — junior vs senior render templates

## AGENTS.md Loading

Always:

```bash
test -f AGENTS.md && cat AGENTS.md
test -f .agent/AGENTS.md && cat .agent/AGENTS.md
```

Parse rule blocks. Where this skill and AGENTS.md disagree, AGENTS.md wins. Cite the AGENTS.md line in the finding: `(AGENTS.md §<section>)`.

## Output Contract

Single markdown document, sections in fixed order:

1. **Summary** — target, base SHA, head SHA, files changed, finding counts by tag.
2. **Important** (🔴) — blocks. Sorted by file path.
3. **Nit** (🟡) — does not block alone.
4. **Pre-existing** (⚪) and **questions** (❓) — collapsible. Pre-existing does not block. Questions never block.
5. **Tooling** — lint summary, tsc summary, test status if available.
6. **Verdict** — `APPROVE` or `BLOCK`. Any Important finding, or any finding that cites an AGENTS.md hard rule, is BLOCK. Otherwise APPROVE.

See [[angular-cop-output-format]] for full templates.

## Boundaries

- Does not write code fixes. Suggestions only.
- Does not run e2e or unit tests by default. If `e2e-runner` is installed, dispatch to that agent for browser flows. If `tdd-guide` is installed, dispatch to that agent for unit tests. Otherwise the main agent loads the `tdd` skill and runs the unit tests inline.
- Does not approve PRs in GitHub/Azure. Author posts the report manually.
- Does not auto-fix lint. Reports counts only.
- If no diff (HEAD == base), exit early with "no changes to review".

From fmflurry/settings-opencode, MIT.

More from fmflurry/settings-opencode