PR Handler

Analyze GitHub PR unresolved feedback and judge validity using GPT-5.2-Codex.

Usage

/pr-feedback 183              → Codex high, PR #183
/pr-feedback 183 deep         → Codex xhigh, PR #183

Argument Parsing

Required:

• •PR number (numeric)

Keywords (reserved):

• •deep → Use xhigh effort instead of high

Parsing Logic

args = split(input)
pr_number = null
effort = "high"

for arg in args:
    if arg == "deep":
        effort = "xhigh"
    elif arg.isdigit():
        pr_number = arg

if pr_number is null:
    error("PR number required")

Result Mapping

Analysis Output

Implementation

Invoke the 17th-pr-handler subagent:

Task(
    description="Analyze PR #{pr_number} feedback ({effort})",
    subagent_type="17th-pr-handler",
    prompt="""
Analyze unresolved feedback for PR #{pr_number}.

Configuration:
- PR number: {pr_number}
- Effort: {effort}  (high or xhigh)
- Project root: {project_root}
""",
    run_in_background=true
)

Workflow After Analysis

1. •Subagent returns analysis report
2. •Main Agent presents to user:
   • •Valid feedbacks → "Apply these?"
   • •Invalid feedbacks → "Decline with reason?"
   • •Needs discussion → "How to respond?"
3. •User decides → Main Agent executes

Output

• •Report: .kkachi/{branch_name}/completed/pr-feedback-{pr_number}-{epoch_timestamp}.md
• •Summary: X valid, Y invalid, Z needs discussion
• •Detailed reasoning for each judgment

Examples

Notes

• •Uses GPT-5.2-Codex with GitHub MCP
• •Read-only analysis (no code modifications)
• •Feedback can be wrong → careful judgment
• •Main Agent handles actual code changes with user
• •Ensure .kkachi/ is in .gitignore