hexmostech/claude-lrc/review

lrc review

Execution policy (strict):

1. Execute commands immediately. Do not ask generic follow-up questions.
2. Never respond with "How can I help" or similar conversational fallback.
3. If no extra user qualifier is provided, default to reviewing staged changes.
4. Use tool calls for command execution before any explanatory prose.

Messaging calibration (strict):

1. Keep status text neutral and concise.
2. Do not frame expected flow as alarming, unexpected, or suspicious.
3. If lrc review --staged commits or pushes, treat that as normal successful behavior.
4. Do not use phrases like "unexpected" or "still failing" unless the flow is truly blocked.

Always ensure the backend is present and review-ready before running review commands.

On Windows use the PowerShell tool:

& "${CLAUDE_SKILL_DIR}/../../scripts/ensure-lrc.ps1" -ForReview

On Unix/macOS use the Bash tool:

bash "${CLAUDE_SKILL_DIR}/../../scripts/ensure-lrc.sh" --for-review

If the command above fails because setup is incomplete:

1. Start setup immediately. Do not ask for an extra yes/no confirmation.

lrc internal claude setup start

1. Tell the user: setup has two steps in order: first complete Hexmos account login in the browser flow, then get a Gemini API key from Google AI Studio; if the browser does not open automatically, complete Hexmos login as needed and open https://aistudio.google.com/ manually; after both steps, paste only the Gemini API key in the next chat reply.
2. When the user replies with the Gemini API key, submit it immediately:

lrc internal claude setup submit-key --key "$USER_MESSAGE"

1. After key submission succeeds, run the ensure command again (platform-appropriate, as above) and then stop.

Then run the mapped review command for this request (lrc review --staged, lrc review --staged --blocking-review, or lrc review --commit HEAD).

If the second ensure step still reports setup required immediately after a successful submit-key, do not run exploratory diagnostics. Proceed directly to the mapped lrc review command once.

Do not run exploratory diagnostics (lrc status, ls, cat, etc.) unless one of the two commands above fails.

Do not prefix canonical commands with source ~/.lrc/env &&.

Do not inspect ~/.lrc.toml with Read/file tools to decide readiness. Use only the canonical ensure script and lrc setup flow.

Then map the user request to the canonical command:

• Default review: lrc review --staged
• If the user explicitly asks for a blocking browser review: lrc review --staged --blocking-review
• If the user asks to review a prior commit instead of the staged tree: lrc review --commit HEAD
• If the user asks for a non-interactive or agent review: lrc review --no-serve --output json --staged
• If the user asks to review a branch or range non-interactively: lrc review --no-serve --output json --range <ref>

Do not use skip or vouch from this skill unless the user explicitly changes intent.