Guide for capturing breakthroughs, lessons learned, and hidden traps from the current session and exporting them into the .opencode/ folder. Use when a task is finished or a hard bug is resolved.
Capture architectural intelligence and "Hidden Traps" discovered during development and convert them into permanent system knowledge.
This skill is triggered when:
.opencode/ instructions were incomplete or led to a mistake.Analyze the delta between the initial failure and the final solution.
py-spy, nvidia-smi) or code pattern fixed it?Map the discovery to the correct .opencode/ category:
| Category | Destination | When to Update |
|---|---|---|
| Domain Intuition | agents/ | New sharding rules, hardware constraints, or framework behaviors. |
| Automation | command/ | Repetitive Git, Bash, or Review workflows found during the session. |
| Risk Mapping | data/ | New correlations between "File Paths" and "Risk Levels/Severities." |
| Procedures | skills/ | Standard Operating Procedures (SOPs) for repeating the task. |
Follow the AReaL Formatting Standard:
print() in hot paths").bash command used to verify the state..opencode/ file.| Component | AReaL Crystallized Lesson | File Reference |
|---|---|---|
| Agent | "Experts must use dp_shard_mod_ep mesh, not dp_shard, to avoid OOM." | agents/archon-expert-example-AReaL.md |
| Command | "Allocate task(model='opus') for engine code but haiku for docs." | command/review-pr-example-AReaL.md |
| Data | "Map parallel_dims.py to CRITICAL risk level." | data/review-pr-change-types-example-AReaL.md |
| Skill | "Verify StateDict on a Meta Device before consuming real GPU time." | skills/add-archon-model-example-AReaL/SKILL.md |
.opencode/ folder.bash commands used for verification.