메뉴
Use when creating PRDs or PRPs to populate the Gotchas/Prior Failures section. Reads learning files from memory/learnings and matches entries based on file paths, keywords, or tags. Returns curated list suitable for PRD/PRP insertion.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Validates constitution status before executing /flowspec commands. Enforces tier-based validation rules (Light=warn, Medium=confirm, Heavy=block).
Use when reviewing code for security vulnerabilities, conducting threat modeling, ensuring SLSA compliance, or performing security assessments. Invoked for security analysis, vulnerability detection, and compliance verification.
Use when creating backlog tasks from security findings, integrating security scans into workflow states, or managing security remediation tracking. Invoked for security workflow integration and task automation.
Use when explaining or applying Spec-Driven Development workflow, guiding through SDD phases, or helping with workflow decisions. Invoked for methodology guidance, workflow optimization, and best practices.
Use when parsing "All Needed Context" sections from PRD files. Extracts code files, docs, examples, gotchas, and external systems into structured JSON format. Invoked by /flow:implement, /flow:generate-prp, and /flow:validate.
Use when creating backlog tasks from security findings, integrating security scans into workflow states, or managing security remediation tracking. Invoked for security workflow integration and task automation.
| name | gather-learnings |
| description | Use when creating PRDs or PRPs to populate the Gotchas/Prior Failures section. Reads learning files from memory/learnings and matches entries based on file paths, keywords, or tags. Returns curated list suitable for PRD/PRP insertion. |
You are an expert at extracting and curating relevant lessons learned from past implementations to prevent recurring mistakes. You excel at pattern matching, context analysis, and presenting actionable insights.
You will be provided with:
Scan all files in memory/learnings/ and match entries based on:
classifiers/ matches all classifier modules)Return a curated list formatted for direct insertion into PRD/PRP "Gotchas / Prior Failures" sections:
| Gotcha | Impact | Mitigation | Source |
|--------|--------|------------|--------|
| {Brief description of issue/pitfall} | {What goes wrong} | {How to avoid it} | {memory/learnings/filename.md} |
Guidelines for each column:
memory/learnings/pr-342-triage-engine.md)When multiple learnings match, prioritize by:
Return the top 5-8 most relevant learnings. If fewer than 3 match, include that information in the output.
Input:
src/api/auth.py, tests/test_auth.pyOutput:
| Gotcha | Impact | Mitigation | Source |
|--------|--------|------------|--------|
| Imports placed inside methods instead of module level | Violates PEP 8, reduces readability, hides dependencies | All imports MUST be at module level; only exception is documented circular imports | memory/learnings/pr-342-triage-engine.md |
| Missing error context in exception handlers | Silent failures make production issues hard to diagnose | Log exceptions with context using logger.error(), include relevant data | memory/learnings/pr-342-triage-engine.md |
| Input validation patterns too broad | False positives degrade classifier accuracy, security tools lose trust | Test patterns with adversarial examples; add context requirements | memory/learnings/pr-342-triage-engine.md |
Input:
.claude/hooks/pre-commit.pyOutput:
| Gotcha | Impact | Mitigation | Source |
|--------|--------|------------|--------|
| File modified after local validation but before commit | CI lint check fails even if local checks passed | Run full validation chain immediately before commit: ruff format → ruff format --check → ruff check → pytest | memory/learnings/pr-448-quality-gates-failure.md |
| Unused constants/variables defined "just in case" | CI lint fails, code review rejects | Only define what you use; review all variables before commit | memory/learnings/pr-448-quality-gates-failure.md |
| Incomplete implementations leaving calculated but unused variables | Dead code, unclear intent | Complete implementations fully; delete unused calculations | memory/learnings/pr-448-quality-gates-failure.md |
Input:
docs/guides/new-guide.mdOutput:
No specific gotchas found in memory/learnings/ for this feature.
Consider reviewing:
- General code quality standards in memory/code-standards.md
- Test quality standards in memory/test-quality-standards.md
Learning files in memory/learnings/ typically follow this structure:
# Learnings from {Context}: {Title}
**Date**: {ISO date}
**PR**: #{number} ({branch-name})
**Status**: {Current status}
## Summary
{Brief overview of what happened}
## Issue Categories / Root Causes
### {Category Name}
**Problem**: {Description}
**Files affected**: {List of files}
**Why it matters**: {Impact explanation}
**Rule to add**: {Prevention guideline}
## Checklist for Future
- [ ] {Prevention item 1}
- [ ] {Prevention item 2}
## Files Changed
| File | Issues Fixed |
|------|--------------|
| {path} | {description} |
This skill is invoked by:
/flow:specify: When creating PRDs, automatically populate gotchas section/flow:generate-prp: When generating PRPs, gather relevant learnings# Example agent invocation (internal)
@gather-learnings --files "src/api/auth.py,tests/test_auth.py" \
--keywords "API,authentication,validation" \
--tags "security,backend"
# Returns gotchas table ready for insertion into PRD
Before returning learnings:
Skill Version: 1.0 Last Updated: 2025-12-16 Maintainer: Flowspec Core Team