[H1][LEARN-SKILL]

Dictum: Delegated synthesis preserves tokens while conveying complete knowledge.

Delegate to skill-summarizer agent. Receive structured skill knowledge.

Parameters:

• $1 — Skill name (required). Fuzzy match to .claude/skills/{name}/.
• $2 — Section (optional). Target specific folder or H2 section.

[1][DELEGATE]

Dictum: Agent reads exhaustively or targeted per request.

Spawn skill-summarizer agent via Task tool.

No section specified: Read ALL files in .claude/skills/$1/, produce complete synthesis.

Section specified: Read only the targeted part:

• references — Read references/*.md files only
• scripts — Read scripts/*.py(or .ts) files only
• templates — Read templates/**/*.md files only
• {H2-label} — Extract specific section from SKILL.md (e.g., validation, frontmatter)

[CRITICAL]:

• [NEVER] Read skill files directly — delegate to agent.

[2][RECEIVE]

Dictum: Dynamic output reflects skill structure.

Agent returns sections based on request:

• Full read: [SKILL], [REFERENCES], [{FOLDER}], [SOURCES]
• Targeted read: [{TARGET}], [SOURCES]

[3][PROCEED]

Dictum: Loaded context enables task execution.

Confirm skill knowledge received. Continue with user task.