upskill

Extract validated PR review findings and update skill checklists to prevent recurring issues.

Updated May 23, 2026
One-click install
npx skills add https://github.com/kiranimmadi2/promptforge-ai --skill upskill-kiranimmadi2
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: upskill
Source: https://github.com/kiranimmadi2/promptforge-ai/tree/main/ai_clients_dart/.agents/skills/upskill
Command: npx skills add https://github.com/kiranimmadi2/promptforge-ai --skill upskill-kiranimmadi2

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Valuable lessons from code reviews often get lost after a PR is merged, causing the same bugs to recur. This Skill closes the feedback loop by mining merged PRs for reviewer findings that the author confirmed as valid, then consolidating them into actionable rules written back into review checklists and implementation pattern files. ## Core Features & Use Cases - Validated Finding Extraction: Scans merged PRs via the GitHub API for review comments the PR author acknowledged with a "Valid." reply, ensuring only confirmed issues are captured. - Pattern Consolidation: Groups recurring findings, generalizes them into actionable rules, and maps each to the right section of core or per-package checklist and pattern files. - Safe Execution Modes: Supports --plan (analysis only), --dry-run (show proposed edits without applying), and full apply mode with user confirmation, plus state tracking via config/state.json to process only new PRs. - Use Case: After a sprint with 15 merged PRs, run the skill to mine review feedback, discover that "missing null-check on API response" appeared 4 times, and add it as a new checklist item so future reviews catch it automatically. ## Quick Start Ask the AI to run the upskill workflow to learn from recent merged PR reviews and update the review checklists with any new validated findings.

Frequently Asked Questions about upskill

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I learn from PR review comments automatically?

Run the upskill workflow, which fetches merged PRs through the GitHub API, finds review comments the PR author acknowledged with a "**Valid.**" reply, and consolidates them into reusable patterns. Confirmed patterns are then written into review checklist and implementation pattern files.

How to update code review checklists from past feedback?

The skill extracts validated findings from merged PRs, groups recurring issues into generalized rules, and classifies each as already covered, an enhancement, or a new addition. After your confirmation, it edits the target REVIEW_CHECKLIST and implementation-patterns markdown files directly.

Can I preview changes before updating skill files?

Yes. Use --plan to run extraction and analysis without editing anything, or --dry-run to see the exact proposed text additions per file and section. Full apply mode also waits for explicit user confirmation before modifying any files.

What tools are required to mine GitHub PR reviews?

The workflow requires git, the GitHub CLI (gh) with an authenticated session, and jq for JSON processing. It also checks the GitHub API rate limit before starting and stops if fewer than 100 requests remain.

Does it reprocess old pull requests on every run?

No. A state.json file records the last checked PR number and run date, so each run only processes PRs merged after that cursor. You can override the starting point with the --from N flag when needed.

What happens when there are many PRs to process?

For 20 or fewer PRs, extraction runs inline in the main conversation. For more than 20 PRs, the workflow delegates extraction to a single subagent with instructions to process the full list and report progress every 10 PRs, avoiding context overflow.