Resolve Review

Process PR review feedback through file-based workflow with user control.

Glossary

See item-model.md for complete schema.

Usage

# Start review session (prompts for mode)
/resolve-review

# Start with specific PR
/resolve-review --pr=123

# Filter by author
/resolve-review --author=copilot

# Resume/apply from previous session
/resolve-review --apply --dir=<tmp>/owner-repo/123/2026-01-16

Arguments

Workflow

Mode Selection

At start, prompt user to choose mode:

How would you like to work through the review items?

[ ] Interactive - I'll present each item, you decide in-session (can defer to file)
[ ] Batch - I'll generate all files, you edit externally, then tell me when ready

Interactive Mode

digraph interactive {
    "Fetch & generate files" [shape=box];
    "Present item" [shape=box];
    "User chooses action" [shape=diamond];
    "Execute action" [shape=box];
    "Read file for response" [shape=box];
    "More items?" [shape=diamond];
    "Summary" [shape=doublecircle];

    "Fetch & generate files" -> "Present item";
    "Present item" -> "User chooses action";
    "User chooses action" -> "Execute action" [label="address/dismiss/etc"];
    "User chooses action" -> "Read file for response" [label="file"];
    "Read file for response" -> "Execute action";
    "Execute action" -> "More items?";
    "More items?" -> "Present item" [label="yes"];
    "More items?" -> "Summary" [label="no"];
}

1. Fetch and generate all item files upfront
2. For each item:
   • Present item with analysis (per terminal-view.md)
   • User chooses: address / dismiss / prior-fix / defer / file
   • If file: User edits the item's .md file, tells CC when ready, CC reads file for response
   • Execute action, move to next
3. Show summary

"file" option: Defers response to the file. User edits response yaml in their editor, returns to CC and says "ready" or "done". CC reads the file and uses whatever response is configured.

Batch Mode

digraph batch {
    "Fetch & generate files" [shape=box];
    "Show file list" [shape=box];
    "User edits files externally" [shape=ellipse];
    "User says ready" [shape=ellipse];
    "Parse all files" [shape=box];
    "Execute all actions" [shape=box];
    "Summary" [shape=doublecircle];

    "Fetch & generate files" -> "Show file list";
    "Show file list" -> "User edits files externally";
    "User edits files externally" -> "User says ready";
    "User says ready" -> "Parse all files";
    "Parse all files" -> "Execute all actions";
    "Execute all actions" -> "Summary";
}

1. Fetch and generate all item files
2. Display file list with directory path
3. User edits files in their editor (modify responses, set enabled: false, delete files)
4. User returns and says "ready" / "apply" / "done"
5. Parse all remaining files via scripts/files-to-items.sh
6. Execute all actions via scripts/apply-items.sh
7. Show summary

File Generation

Both modes generate files upfront:

1. Fetch via scripts/fetch-pr-items.sh
2. Analyze each item and populate analysis section
3. Generate files via scripts/items-to-files.sh

Output directory: <tmp>/<owner>-<repo>/<pr>/<date>/

File format: file-view.md

Actions

Reply Tone

Replies post from user's GitHub account. Keep them:

• Brief (1-2 sentences)
• Direct and factual
• No AI pleasantries
• No hedging

Examples:

• "Fixed in abc123."
• "Moved early return above the selector call."
• "Out of scope for this PR, tracking in HEA-1234."

Critical Rules

1. User controls timing - Generate creates files, Apply executes (separate steps)
2. File is source of truth - User edits override CC analysis
3. enabled: false skips - Don't delete files to preserve context
4. Defer = thread stays open - No GitHub changes

Reference

• item-model.md - Item schema (source of truth)
• file-view.md - Markdown file format
• terminal-view.md - Terminal display format
• api-reference.md - GitHub GraphQL API
• scripts/ - Bash scripts for fetch/serialize/apply