chore: update review code skill output and tone

[nabinchha]

Summary

• Remove the Overview metadata table from review output — the review now starts with a cordial thank-you and goes straight into the Summary
• Add a Tone section with guidelines for cordial, collaborative reviews (supportive teammate, not gatekeeper)
• Tie the Verdict strictly to the highest-severity finding: "Ship it (with nits)" is limited to Suggestions only; any Warning or Critical forces "Needs changes"
• Soften severity section headers ("Let's fix these before merge", "Worth addressing", "Take it or leave it")
• Add Step 7 to post the review directly as a GitHub PR comment via gh pr comment
• Omit empty severity sections from output

Example review generated with the updated skill

Closes #476

[greptile-apps]

Greptile Summary

This PR refines the review-code skill's output format and tone to produce more collaborative, readable PR reviews. The implementation closely matches the stated intent.

Key changes:

• Removed Overview tables — the metadata tables for PR/Branch mode are dropped; reviews now open with a brief thank-you and go straight to the Summary section
• Added ## Tone section — placed immediately before Step 6 so tone guidance is co-located with the output template; encourages "supportive teammate" framing over gatekeeper language
• Tightened Verdict logic — "Ship it (with nits)" is now strictly limited to Suggestion-only reviews; any Warning or Critical forces "Needs changes", removing the previous ambiguity
• Softened severity headers — "Let's fix these before merge", "Worth addressing", "Take it or leave it" match the new collaborative tone
• AI signature added — PR-mode reviews end with a disclaimer line distinguishing agent output from human reviews
• Step 7 updated with confirmation gate — the agent now displays the review and explicitly asks for user confirmation before running gh pr comment, addressing the safety concern raised in prior review rounds
• "Next steps (optional)" section removed — keeps the output leaner
• One minor logical ambiguity remains in the Verdict section: the severity hierarchy (Critical → Warning → Suggestion) doesn't specify precedence when both Warning-level code findings and a "Needs discussion" design question exist simultaneously in the same review

Confidence Score: 5/5

• Safe to merge — changes are documentation/skill-prompt only with no runtime code affected.
• All P0/P1 concerns from prior rounds (Tone placement, confirmation gate before posting) are resolved in this revision. The single remaining note is a P2 ambiguity in the Verdict selection logic that would only surface in an edge-case scenario; it does not block merge.
• No files require special attention.

Important Files Changed

Filename	Overview
.agents/skills/review-code/SKILL.md	Review skill updated: removes Overview metadata table, adds Tone section before Step 6, tightens Verdict criteria to severity hierarchy, softens severity headers, adds AI signature, and adds confirmation-gated Step 7 for posting. One minor logical ambiguity in Verdict selection when both Warning findings and "Needs discussion" issues coexist.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Review Complete] --> B{Any findings?}
    B -- No --> V1[🟢 Ship it]
    B -- Yes --> C{Design/arch\nquestions?}
    C -- Yes --> V4[💬 Needs discussion]
    C -- No --> D{Highest severity?}
    D -- Critical or Warning --> V3[🔴 Needs changes\nList blocking items]
    D -- Suggestions only --> V2[🟡 Ship it with nits]
    V3 --> E[Step 7: Display review\n+ ask user to confirm\nbefore posting]
    V2 --> E
    V1 --> E
    V4 --> E
    E --> F{User confirms?}
    F -- Yes --> G[gh pr comment <number>\n--body-file /tmp/review-<n>.md]
    F -- No --> H[Done — review saved\nto /tmp/review-<n>.md]

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: .agents/skills/review-code/SKILL.md
Line: 244-249

Comment:
**Verdict selection ambiguous when both Warnings and "Needs discussion" apply**

The instruction "Choose the verdict that matches the **highest severity** finding" establishes a clear severity ladder (Critical → Warning → Suggestion), but `Needs discussion` isn't a severity level — it's triggered by a qualitatively different condition (unresolved architectural/design questions). When a review contains both a Warning-level code finding *and* a blocking design question, the agent has no rule to decide which verdict wins.

Consider making the precedence explicit, for example:

```suggestion
Choose the verdict that matches the **highest severity** finding in the review. If there are both code findings and unresolved design questions, prefer **Needs discussion** since those questions must be settled before code-level feedback can be finalized:

- **Ship it** — No findings. Ready to merge as-is.
- **Ship it (with nits)** — Only Suggestions (see above — style improvements, simplifications, or optional enhancements). Nothing blocking.
- **Needs changes** — Any Critical or Warning findings. List the items that must be addressed before merge.
- **Needs discussion** — Architectural or design questions that need team input before a decision can be made. Takes precedence over Needs changes when the design itself is unsettled.
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (5): Last reviewed commit: "Merge branch 'main' into nmulepati/chore..." | Re-trigger Greptile}