self-review

Forge reviews its own skill updates, review history, and research outputs before publishing them.

SKILL.md

GitHub

Forge Self-Review Protocol

When to Self-Review

  • After updating any SKILL.md file — review before saving
  • After completing a heartbeat research cycle — review findings before sharing
  • After adding a new anti-pattern to any skill — verify it's accurate
  • Weekly: review the last 10 entries in review-history/ for quality

How to Self-Review a Skill Update

  1. Read the current skill file
  2. Ask: Is this accurate? Is it based on source code or docs?
  3. Ask: Is there anything outdated or contradicted by the actual framework source?
  4. Ask: Would a senior engineer following this produce better code?
  5. If all yes → save. If any no → revise first.

How to Self-Review Review History

Every 5 reviews, check review-history/ for:

  • Did I miss anything that QA caught later? → Add to relevant skill as anti-pattern
  • Am I flagging the same issue repeatedly? → Send Maks a direct tip proactively
  • Did I block something that didn't need blocking? → Calibrate severity downward

How to Self-Review Research Findings

Before sharing a CVE or best practice:

  • Verify CVE is real (check nvd.nist.gov or GitHub advisory database)
  • Verify it affects our stack version specifically
  • Don't share framework updates that don't affect our stack
  • Rate confidence: HIGH (confirmed in source) / MEDIUM (docs) / LOW (community report)

Quality Bar for Forge's Own Output

  • Every review must have at least one "What's Done Well" item
  • No vague feedback — always file:line:issue:fix format
  • P0/P1 blocks must have exact fix instructions, not just "fix this"
  • Research alerts must include: source URL, affected version, specific fix

Changelog

  • 2026-03-20: Created