fugu-verify

name	fugu-verify
description	Fugu-lite Verifier for the current changes — inspects git status/diff, detects the package manager and scripts, runs typecheck/lint/test/build where available, scans for secrets, runs a Fusion-lite review, and produces an approve / request_changes / needs_human_review decision. Does NOT implement new changes unless the user explicitly asks. Use to check work before committing.
disable-model-invocation	true
allowed-tools	Read, Grep, Glob, Task, Bash(git status:*), Bash(git diff:*), Bash(git show:*), Bash(git log:*), Bash(npm test:*), Bash(npm run:*), Bash(pnpm test:*), Bash(pnpm run:*), Bash(yarn test:*), Bash(yarn run:*), Bash(npx tsc:*), Bash(find:*), Bash(ls:*), Bash(cat package.json)

Fugu-lite Verify (Verifier)

Verify the current changes without implementing new ones. Do not modify files unless the user explicitly asks. Apply the rubric at ${CLAUDE_PLUGIN_ROOT}/skills/fugu-verify/rubrics/verification-rubric.md.

Input

$ARGUMENTS may narrow the scope (a path or "the last commit"). If empty, verify the current uncommitted working-tree changes.

Process

You may delegate to the fugu-verifier subagent via the Task tool, or run inline.

1. Inspect git status --short and git diff — know exactly what changed.
2. Detect the project (you may run ${CLAUDE_PLUGIN_ROOT}/scripts/detect-project.sh):
   • pnpm-lock.yaml → pnpm, yarn.lock → yarn, package-lock.json → npm.
   • Read package.json scripts: test, typecheck, lint, build.
3. Run reasonable checks that exist (you may run ${CLAUDE_PLUGIN_ROOT}/scripts/run-checks.sh). Do not install dependencies; skip missing tools and say so.
4. Secret scan the diff (${CLAUDE_PLUGIN_ROOT}/scripts/secret-scan.sh); never print secret values.
5. Fusion-lite review of the diff — call /fusion-review (or the reviewer subagents + fusion-judge).
6. Distinguish failures caused by this change from pre-existing/unrelated ones, with evidence.

Output

A short report plus the Fusion-style decision JSON:

{
  "decision": "approve | request_changes | needs_human_review",
  "severity": "none | low | medium | high | critical",
  "summary": "",
  "issues": [],
  "confidence": "low | medium | high",
  "follow_up_commands": []
}

Plus, in prose: the checks you ran (command → result), pre-existing vs. caused-by-change failures, remaining risks, and whether it is safe to commit. Do not commit or push.