init

Copydesk Init

This skill bootstraps the copydesk user-data directory and walks you through extracting a voice register. It's idempotent — running it again is how you add a new register or refresh templates after a plugin update.

Phase 1: Bootstrap the data directory

1. Always ensure the required subdirectories exist. mkdir -p is idempotent — these calls are safe to run on every invocation, and they protect against partial-init states (e.g., the root exists but registers/ was deleted, or the data root was created by a prior plugin version without all its subdirs):

   mkdir -p ~/.claude/data/copydesk/registers
   mkdir -p ~/.claude/data/copydesk/learning/snapshots
   

2. For every file under ${CLAUDE_PLUGIN_ROOT}/template-data/, check whether the corresponding target under ~/.claude/data/copydesk/ exists. If the target is missing, copy the template file there. Never overwrite a file that already exists at the target. This pattern lets plugin updates ship new template files (which appear at the data path on next init invocation) without ever touching existing user data.

Phase 2: Detect state and route

After Phase 1, list ~/.claude/data/copydesk/registers/ excluding register-template.md.

• Fresh install (zero populated registers): continue to Phase 3 — first-time extraction walkthrough.
• At least one populated register: ask the user "You already have these registers: [list]. Do you want to (1) add a new register, or (2) re-extract an existing register?" Default to (1). If (2), prompt for which register; treat it as a fresh extraction that will overwrite the chosen register file when complete.

Phase 3: Extraction walkthrough

The current process follows the existing setup/ reference documents. (The user has flagged that this entire phase will be replaced by the planned extraction/learning rework; the bootstrap in Phases 1-2 is the stable structural piece.)

Step 3.1: Identify the register

Ask the user what kind of writing this register is for (e.g., advocacy, personal essays, technical documentation, dystopian fiction). Derive a short kebab-case name (e.g., advocacy, personal, tech-docs, dystopian-fiction) that becomes the register's filename. Confirm with the user.

Also ask: "What writing contexts should activate this register?" Capture the list — it becomes the triggers: array in the register's frontmatter.

Step 3.2: Collect samples

Reference ${CLAUDE_PLUGIN_ROOT}/setup/sample-collection.md for the format requirements. Ask the user for 10-20 samples of their own writing in this register, plus 10 baseline samples (Claude-default outputs on similar topics).

If the user doesn't have baselines ready, walk them through generating a batch: open a fresh Claude conversation (no system prompt, no special instructions), ask it to write 10 short pieces (150-300 words each) on topics similar to their writing samples using prompts like "Write a short comment about [topic]." Save the outputs as the baseline (P1) corpus.

Step 3.3: Pass-1 extraction

Read the prompt at ${CLAUDE_PLUGIN_ROOT}/setup/pass-1-prompt.md. Dispatch a Sonnet agent with that prompt as system context, filling in:

• the P1 section with the user's baseline samples
• the P2 section with the user's own writing samples

Save the agent's output to ~/.claude/data/copydesk/learning/extraction-artifacts/<register-name>/pass-1-output.md (create the directory if needed).

Step 3.4: Pass-2 extraction

Read the prompt at ${CLAUDE_PLUGIN_ROOT}/setup/pass-2-prompt.md. Dispatch a second Sonnet agent with that prompt, passing the pass-1 output from Step 3.3. Save the result to ~/.claude/data/copydesk/learning/extraction-artifacts/<register-name>/pass-2-output.md.

Step 3.5: Write the register file

Convert the pass-2 output into a register file using the structure in ${CLAUDE_PLUGIN_ROOT}/template-data/registers/register-template.md. Important:

1. Write YAML frontmatter at the top with the triggers: array from Step 3.1.
2. Place the pass-2 voice feature description (Vocabulary / Sentence Structure / Rhetorical Techniques / Voice Qualities sections) in the body, following the template's section order.
3. Save the result to ~/.claude/data/copydesk/registers/<register-name>.md.

Example finished register frontmatter:

---
triggers:
  - personal essays
  - blog posts
  - reflective writing
---

# Personal Register
...

Step 3.6: Optional — brief-stripping setup

Ask the user whether they want brief-stripping support for this register. If yes, walk through ${CLAUDE_PLUGIN_ROOT}/setup/brief-stripping-guide.md. (This is optional; skip if the user is unsure.)

Phase 4: Confirm and exit

Tell the user:

Register <register-name> is ready at ~/.claude/data/copydesk/registers/<register-name>.md with these triggers: [list]. Try it now: invoke /copydesk with writing that matches one of the trigger contexts.

Stop. Generation belongs to /copydesk; this skill's job is done.

Re-running

This skill is idempotent. Re-invocation:

• Phase 1 re-checks the data directory; missing template files are restored, existing files left alone.
• Phase 2 routes based on what's already configured.
• Phase 3 produces a new register (or overwrites an existing one if the user chose re-extraction).

Nothing in this skill writes to the plugin install path. All persistent state lives at ~/.claude/data/copydesk/.