learn-along

Learn Along - Conversational Language Learning

Augment everyday software engineering conversations with language coaching for intermediate+ learners.

When to Use This Skill

• User runs /learn-along or /learn-along <target-language>
• User mentions wanting to practice a language during Claude Code sessions
• User says "learn along", "language coach", or similar

Activation

When the user invokes /learn-along:

1. Parse arguments:
   • /learn-along en → target: English, native: auto-detect
   • /learn-along from:ja to:en → native: Japanese, target: English
   • /learn-along (no args) → target: English, native: auto-detect
   • /learn-along off → deactivate
2. Confirm activation with a brief message stating the configured languages.
3. From this point forward, after every response, invoke the learn-along subagent.

Auto-Detection of Native Language

When native language is not explicitly specified:

1. Detect from the user's first non-target-language message after activation.
2. If the user only writes in the target language, ask once to confirm their native language.
3. Once detected, confirm: "Native language detected: [X]. Use /learn-along from:<native> to:<target> to adjust."

Deactivation

When the user runs /learn-along off or says "stop language coaching":

1. Stop invoking the learn-along subagent.
2. Confirm deactivation.

State

• Learn-along mode is active for the current conversation session only.
• If context becomes unclear after a long conversation, briefly check: "Learn Along is still active (target: English). Say /learn-along off to stop."
• Activation does not persist across sessions. The user must re-activate in new sessions.

Execution Flow (Per Exchange)

IMPORTANT: Always complete the user's technical request fully before adding coaching.

Step 1: Respond Normally

Process the user's request and provide your full technical response as usual.

Step 2: Invoke the learn-along Subagent

Use the Task tool with subagent_type: "learn-along" and model: "haiku". The subagent will classify the input and produce appropriate coaching output.

Provide in the prompt:

• The user's original message
• A brief summary of your response (not the full text if lengthy)
• The native language and target language
• The domain context (e.g., "debugging test failures", "PR review", "architecture discussion")

Example invocation:

Task tool call:
  subagent_type: "learn-along"
  model: "haiku"
  prompt: |
    Native: Japanese, Target: English
    Domain: debugging test failures

    User message:
    このPRのテストが落ちてるから原因を調べて

    Assistant response summary:
    Investigated test failures, found mock setup issue in auth_test.rb, fixed by updating the expected arguments.

Step 3: Append the Coaching Output

Append the subagent's output at the end of your response after a --- separator.

Output Format

---

**Learn Along** (target: English)

[Subagent output here - varies by category]

Guidelines

• Never let coaching interfere with the primary task. Technical response comes first.
• Keep it brief. 3-8 lines for the coaching section.
• Intermediate+ level. No basic grammar explanations. Focus on idiomatic usage, register, nuance, collocations, and technical vocabulary.
• Software engineering context. Prioritize expressions used in PRs, Slack, standups, and technical docs.
• Natural, not textbook. Favor how native speakers actually talk in professional settings.