| name | sigma-lessons |
| description | Recall sigma's past ratcheted lessons before implementing or verifying a task, so previous mistakes are not repeated. Use when about to write or review code in any sigma domain (classic-ml, deep-learning, nlp, rl, data-analysis, data-engineering, ai-agent-engineering, mlops, llm-engineering), when starting a loop task, or when the user asks "what have we learned" / "avoid past mistakes". Lessons are written by the loop on failure and by /sigma-learn-lesson; this skill reads them back.
|
| origin | sigma |
sigma-lessons
sigma accumulates lessons from failures (the loop ratchets them) and from the
human (/sigma-learn-lesson). Each lesson is a skills/<slug>/SKILL.md tagged
with metadata: domain:. This skill reads them back so a future task applies
them instead of repeating the mistake.
When to use
- Before implementing or verifying a task in a known domain — pull the lessons
for that domain first.
- When the user asks to recall what was learned, or to avoid prior mistakes.
Workflow
- Identify the domain of the current task (one of the 9 sigma domains; if a
task line is annotated
(domain), use that).
- Read the lessons for that domain: scan
skills/**/SKILL.md for files whose
frontmatter domain: matches. Each has a **Lesson (ratcheted):** line and a
**How to apply:** line — those are the actionable parts.
- Apply them as constraints while implementing or as extra checks while
verifying. Treat a lesson as "do not repeat this mistake."
- If a lesson looks stale or wrong, do NOT silently delete it — flag it for the
human (sigma never auto-resolves lessons; see
skills/CONTRADICTIONS.md).
Notes
- The CLI loop does this automatically (
cli/skills_recall.py builds the recall
block and injects it into the implement + verify prompts). This skill is the
in-session equivalent for slash-command work.
- Lessons without a
domain: (vendor skills, sigma-present, sigma-domains) are
NOT lessons — ignore them here.
