This skill should be used when the user redirects, corrects, or refines agent output — "no, actually", "that's not right", "you missed something". Captures correction moments to improve Forge workflows over time. Always active during feature work; log corrections as they occur.
From forgenpx claudepluginhub flox/forge-plugin --plugin forgeThis skill uses the workspace's default tool permissions.
Guides Next.js Cache Components and Partial Prerendering (PPR) with cacheComponents enabled. Implements 'use cache', cacheLife(), cacheTag(), revalidateTag(), static/dynamic optimization, and cache debugging.
Migrates code, prompts, and API calls from Claude Sonnet 4.0/4.5 or Opus 4.1 to Opus 4.5, updating model strings on Anthropic, AWS, GCP, Azure platforms.
Details PluginEval's skill quality evaluation: 3 layers (static, LLM judge), 10 dimensions, rubrics, formulas, anti-patterns, badges. Use to interpret scores, improve triggering, calibrate thresholds.
All Forge agents should track "correction moments" — instances where the user redirects, corrects, or refines the agent's output. These are valuable signals for workflow improvement.
Each note includes a status marker for incremental processing
by /forge:improve.
## YYYY-MM-DD: {Phase}
### Note {N}: {Brief Title}
<!-- STATUS: UNPROCESSED -->
**Context:** ...
**What happened:** ...
**Recommendation:** ...
---
Status values:
UNPROCESSED — New note, not yet reviewed by auditPROCESSED: {date} | Applied: {file} | PR: #{pr} —
Change was madePROCESSED: {date} | Skipped: {reason} — Decided not to
actPROCESSED: {date} | Deferred: #{issue} — Created ticket
for laterNotes without any status marker are treated as UNPROCESSED
(for backwards compatibility with existing notes).
A correction occurs when the user:
Track corrections when they indicate:
Don't track minor stylistic preferences or typo fixes.
When you detect a correction, append to
{feature_path}/artifacts/retrospective-notes.md:
## {date}: {Phase}
### Note {N}: {Brief Title}
<!-- STATUS: UNPROCESSED -->
**Agent:** {agent name}
**What happened:**
{Brief description of what the agent did or suggested}
**User correction:**
{What the user said or did to correct it}
**Root cause:**
{Why did the agent get it wrong? Missing context? Bad
assumption? Template gap?}
**Recommendation:**
{What change might prevent this in the future?}
---
If artifacts/retrospective-notes.md doesn't exist, create
it with this header:
# Retrospective Notes: {feature_name}
Correction moments captured during feature development.
These inform workflow improvements.
---
## 2026-01-09: Requirements
### Note 1: Missing infrastructure dependencies section
<!-- STATUS: UNPROCESSED -->
**Agent:** requirements-gatherer
**What happened:**
Generated requirements without an "Infrastructure
Dependencies" section.
**User correction:**
"We need a section for infrastructure dependencies —
this feature requires specific server capabilities."
**Root cause:**
Requirements template doesn't have an infrastructure
dependencies section.
**Recommendation:**
Add "Infrastructure Dependencies" section to requirements
template.
---
## 2026-01-09: Design
### Note 2: Incorrect component placement in architecture
<!-- STATUS: PROCESSED: 2026-01-15 | Applied: context/product.md | PR: #42 -->
**Agent:** designer
**What happened:**
Suggested adding feature support directly to the wrong
component.
**User correction:**
"This should go through the data layer, not the API layer."
**Root cause:**
Agent didn't understand the separation between API and
data layer responsibilities.
**Recommendation:**
Add data layer vs API layer responsibility breakdown to
context/product.md.
---
## 2026-01-09: Design
### Note 3: Incorrect project impact assessment
<!-- STATUS: PROCESSED: 2026-01-16 | Deferred: #55 -->
**Agent:** designer
**What happened:**
Listed a service as affected for a client-only change.
**User correction:**
"This doesn't touch that service at all, it's entirely
client-side."
**Root cause:**
Agent assumed all environment changes require service
updates.
**Recommendation:**
Update designer to distinguish client-only vs
service-affecting changes.
---
Agents should:
Example response when detecting a correction:
"Got it, I'll update my approach. [Noting this for the retrospective — the {X} assumption wasn't correct for this case.]"
The bracketed note signals transparency without being disruptive.