Learn

Objective

Create a topic-specific skill folder that captures durable, source-backed knowledge other agents can reuse later.

Inputs

• Capture the topic from the user request.
• Capture the output root. Default to skills/ in the current workspace if the user does not specify another path.
• Capture depth as quick (5-8 sources) or deep (10-20 sources). Default to deep for technical topics.

Workflow

1. Define scope and skill name.

• Normalize the topic into a slug and append -kb.
• Example: Amazon Kinesis -> kinesis-kb.
• Keep scope narrow enough to be useful (service-level or domain-level), not broad like "cloud".

2. Research with high-quality sources.

• Use current, authoritative sources first: official docs, standards, primary vendor docs.
• Add secondary sources only for context.
• Record URLs and access dates while researching.
• Read references/research-rubric.md before collecting sources.

3. Scaffold the target skill.

• Run:

python3 scripts/scaffold_topic_kb.py "TOPIC" --out skills

• Use --dry-run first when path or naming is uncertain.

4. Fill the generated knowledge sections.

• Edit the generated SKILL.md and replace placeholders with concise, practical guidance.
• Prefer operationally useful content: architecture patterns, pitfalls, troubleshooting, decision criteria, and API caveats.
• Keep statements grounded in cited sources. If uncertain, mark uncertainty explicitly.

5. Add and verify references.

• Populate references/sources.md with the source list used to build the KB.
• Include publication date when available and access date for each source.
• Add a "Last verified" date in generated SKILL.md.

6. Validate and report.

• Validate generated skills with:

python3 "${CODEX_HOME:-$HOME/.codex}/skills/.system/skill-creator/scripts/quick_validate.py" skills/TOPIC-kb

• Report created files, validation status, and key coverage areas.

Output Contract

Create this structure:

<output-root>/<topic>-kb/
  SKILL.md
  agents/openai.yaml
  references/sources.md

Generated skill quality bar:

• Include enough detail for direct problem-solving, not just definitions.
• Include pitfalls and troubleshooting guidance.
• Include source-backed recommendations, not invented claims.
• Keep content compact and scan-friendly.

Resources

• scripts/scaffold_topic_kb.py: Create a topic-kb skill skeleton with valid frontmatter and agent metadata.
• references/research-rubric.md: Apply source-quality and synthesis rules while researching.

Example Invocation

• "Use the learn skill on Amazon Kinesis and create kinesis-kb in skills/."
• "Research OpenTelemetry collectors deeply and produce opentelemetry-kb with citations."