I want my codebase scanned for convention violations

A full architecture pass that discovers the conventions in your repo and flags where the code breaks them.

A full architecture pass that discovers the conventions already in your repo and flags every place the code breaks them.

The prompt

Paste this into Claude Code
$ Sentinel scan this repo, then guard the staged changes against whatever conventions you find.

What the Algorithm does

The Algorithm chains two Sentinel workflows back to back. Scan reads the repo once, infers conventions with a Sonnet pass, and writes them to the MemPalace knowledge graph. Guard then checks the current working tree against the freshly-discovered rules.

What you'll see
$ ♻︎ Entering the DOS ALGORITHM… (v0.0.1) ═════════════🗒️ TASK: Scan and guard this codebase━━━ 👁️ OBSERVE ━━━ 1/7🔎 REVERSE ENGINEERING:• Wants: discovered conventions, violation report• Not wanted: generic linter output, opinions from nowhere💪🏼 EFFORT LEVEL: Standard | Two deterministic workflows chained🏹 CAPABILITIES SELECTED:• Sentinel (Scan) — discover conventions• Sentinel (Guard) — enforce against staged changes• MemPalace — persist conventions as KG triples━━━ ⚡ EXECUTE ━━━ 5/7Scan runs a single Sonnet call in ~45s, emits 18 conventions to the KG. Guard reads staged diff, finds 3 violations in 12s.━━━ ✅ VERIFY ━━━ 6/7Each violation cites the convention it broke and the file:line it appears on.✓ Done. 18 conventions captured, 3 violations flagged.

What you get back

Two artifacts. First, a convention report: the rules Sentinel inferred from your actual code, grouped by domain. Second, a violation list: specific file:line callouts that break those rules, each with a one-line fix suggestion. The conventions stay in MemPalace so every future DOS session starts already knowing them.

Variations

  • Scan the repo but only guard files I'm about to commit
  • Review the entire current branch, not just staged changes
  • Show me the top 5 conventions and stop there
  • Evolve the conventions, I just changed our naming style

Was this page helpful?