Install
$ agentstack add skill-carlkibler-agent-skills-second-opinions ✓ scanned · ✓ verified, works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
Verified badge
Passed review? Show it. Paste this badge into your README, it links to the public security report.
Reliability & compatibility
Declared compatibility
Compatibility is declared by the source manifest. End-to-end runtime verification is coming, see below.
We're building live execution health for every listing: tool-call success rate, median latency, uptime, and last-checked timestamps, measured, not self-reported. It isn't live yet, so we don't show numbers we can't stand behind.
How agent discovery & health will work →About
Second Opinions
Use an independent provider before committing meaningful changes. The reviewer is a collaborator, not an authority. A timeout or provider failure is a failed review, never approval.
When to use
Mandatory for complex multi-file changes, security or PHI work, performance-critical code, design decisions with real tradeoffs, and work that has consumed more than two hours. Skip trivial fixes.
One bounded entrypoint
Use the installed review-watch command. It has a 10-minute total deadline across both attempts, prints 15-second heartbeats, captures output, and terminates the entire review process tree on timeout. Do not invoke bare codex review, claude -p, codex exec, or an unbounded agent call for review. A naked review-watch invocation prints copy-pasteable examples instead of failing with a grammar error. --agent prints the same contract for a first-pass agent.
Tell it which provider is running the current task. It tries the opposite company first, then the same company if the first provider is unavailable, exhausted, or returns an error while time remains:
# Running under Codex: Claude first, Codex fallback
review-watch --current-provider codex --worktree "$PWD"
# Running under Claude: Codex first, Claude fallback
review-watch --current-provider claude --worktree "$PWD"
The script uses Claude Opus and Codex gpt-5.6-sol by default. It accepts common aliases such as openai/gpt, anthropic/opus, --against for --base, --diff for --range, comma-separated SHAs, bare worktree/ref/range positional arguments, and durations such as 10m, 600s, or 1h. Override only when needed: --claude-model MODEL, --codex-model MODEL, --timeout SEC, or --heartbeat SEC. Use --json when a caller needs a designed machine-readable receipt; progress and provider diagnostics stay on stderr so stdout remains valid JSON.
Scope is explicit
A bare SHA is not enough context for a branch review. Prefer the complete branch diff:
review-watch --current-provider codex --worktree /absolute/path/to/worktree --base main
On a topic branch with no scope flags, the script automatically reviews the branch against its main merge-base. Use these forms when the scope is narrower or spans selected history:
review-watch --current-provider codex --worktree "$PWD" --commit 470907dae
review-watch --current-provider codex --worktree "$PWD" --commit SHA1 --commit SHA2
review-watch --current-provider codex --worktree "$PWD" --range OLD..NEW
--worktree changes the checkout the provider inspects. The prompt always includes its absolute path, branch, status snapshot, exact scope, and the instruction to inspect the complete diff. The provider must not edit files or perform live mutations. Output restates canonical WORKTREE, BRANCH, SCOPE, provider order, deadline, and ends with runnable Next commands plus REVIEW_STATUS, REVIEW_PROVIDER, and REVIEW_EXIT keys.
Review lens
Check correctness and regressions first, then security/PHI, real outbound effects, performance, and maintainability. Return PASS or NEEDS CHANGES. Every finding needs severity, path, line, concrete evidence, and the smallest fix. If the scope is unavailable or ambiguous, fail closed.
Record the provider, exact command, scope, exit status, and findings in the review note or merge receipt. A fallback is a real review; report that the primary provider failed and which fallback ran.
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: carlkibler
- Source: carlkibler/agent-skills
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet, be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.