<tool_restrictions>

MANDATORY Tool Restrictions

BANNED TOOLS — calling these is a skill violation:

EnterPlanMode — BANNED. This is a procedural skill. Execute it directly.
ExitPlanMode — BANNED. You are never in plan mode. </tool_restrictions>

Verify Workflow

Run sequential verification checks. Stop on critical failures.

Step 0: Detect Tooling

Before running checks, detect the project's tooling:

Package manager:

pnpm-lock.yaml → pnpm
bun.lockb or bun.lock → bun
yarn.lock → yarn
package-lock.json → npm

Linter:

biome.json or biome.jsonc → [pm] biome check .
.eslintrc* or eslint.config.* → [pm] eslint .

Test framework:

vitest.config.* → [pm] vitest run
jest.config.* → [pm] jest

Type checker:

tsconfig.json → [pm] tsc --noEmit

Build:

Check package.json scripts for build → [pm] run build

Step 1: Parse Mode

$ARGUMENTS determines the mode:

Argument	Mode	Checks
`quick`	Quick	Build + types only
`full` or (none)	Full	All checks
`pre-commit`	Pre-commit	Build + types + lint + debug logs (skip tests)
`pre-pr`	Pre-PR	All checks + search for hardcoded secrets

Step 2: Run Checks (in order)

Execute each check sequentially. If a check fails critically, report it and stop.

2a. Build Check

[pm] run build

If FAIL: Report errors and STOP — nothing else matters if it doesn't build
If PASS: Continue

2b. Type Check

[pm] tsc --noEmit

Report error count and locations (file:line)
Continue even if there are errors (report them in summary)

2c. Lint Check

[pm] biome check .
# or
[pm] eslint .

Try auto-fix first: [pm] biome check --write .
Report remaining issues
Continue

2d. Test Suite (skip in `quick` and `pre-commit` modes)

[pm] vitest run
# or
[pm] jest

Report pass/fail counts
Report coverage percentage if available
Continue

2e. Debug Log Audit

Search for leftover debug statements in source files (not test files):

Use Grep tool: Pattern console\.(log|debug|dir|table) in src/ or app/ directories, excluding *.test.* and *.spec.* files

Report locations
Note: console.warn and console.error are intentional — don't flag those

2f. Git Status

git status --short
git diff --stat

Report uncommitted changes count
Report staged vs unstaged

2g. Secrets Scan (pre-pr mode only)

Use Grep tool: Search for patterns that suggest hardcoded secrets:

sk_live_, sk_test_, pk_live_, pk_test_ (Stripe)
AKIA (AWS access keys)
ghp_, gho_, ghs_ (GitHub tokens)
xoxb-, xoxp- (Slack tokens)
Strings assigned to variables named *_KEY, *_SECRET, *_TOKEN that look like real values (not env var references)

Exclude: .env*, *.example, *.md, node_modules/

Step 3: Summary Report

VERIFICATION: [PASS / FAIL]

Build:      [OK / FAIL]
Types:      [OK / X errors]
Lint:       [OK / X issues]
Tests:      [X/Y passed / SKIPPED]
Debug logs: [OK / X found]
Git:        [clean / X uncommitted]
Secrets:    [OK / X found / SKIPPED]

Ready for PR: [YES / NO]

If any check failed, list the specific issues below the table with file:line references and brief fix suggestions.

Edge Cases

No build script: Skip build check, note it in summary
No test framework: Skip tests, note it in summary
No TypeScript: Skip type check, note it in summary
Monorepo: If turbo.json or root pnpm-workspace.yaml exists, run checks at the workspace level (turbo build, turbo typecheck)

What Verify Does NOT Do

Fix issues (use the fixer agent for that)
Run E2E tests (use /arc:testing for that)
Block anything — it reports, the developer decides
Modify any files

verify