draft-import

name	draft-import
description	Import markdown content from a local directory or private GitHub repo into the active workspace. Analyzes source files, maps content to Draft dimensions, and stages the result as a proposal for review. Nothing is applied until the user accepts.

/draft:import — Import Context from an Existing Source

Imports existing knowledge into your Draft workspace. Reads markdown files from the source, maps them to context dimensions using AI judgment, and stages a reviewable proposal — nothing is written to your workspace until you accept.

---

Usage

/draft:import <source>

Where <source> is:

• A local path: ~/notes, /Users/you/obsidian-vault, ./docs
• A GitHub repo: owner/repo or https://github.com/owner/repo

---

Step 1: Detect source type

If <source> starts with ~/, /, or ./ → local path. Otherwise → GitHub repo.

---

Step 2: Fetch the content

Local path

Expand ~ if present. Verify the directory exists — if not, tell the user and stop.

Use Glob to collect all .md files under the path (skip .git/, node_modules/). Read each file. If there are more than 50 files, ask the user to confirm before proceeding.

GitHub repo

Prerequisites — check both before proceeding:

command -v gh >/dev/null 2>&1

If gh is not installed: "Install the GitHub CLI first: brew install gh" — stop.

gh auth status >/dev/null 2>&1

If not authenticated: "Run gh auth login --web then retry /draft:import." — stop.

Clone to temp dir (separate-clone pattern):

IMPORT_DIR=$(mktemp -d /tmp/draft-import-XXXXXX)
gh repo clone <repo> "$IMPORT_DIR" -- --depth 1 --quiet 2>&1

If the clone fails (repo not found, no access): tell the user and stop. The trap will clean up.

Collect .md files from $IMPORT_DIR (same as local). After reading all files, delete the temp dir:

rm -rf "$IMPORT_DIR"

---

Step 3: Analyze and map

Read all collected files. For each file, use judgment to map it to a Draft dimension:

Signals to use:

• Parent directory name matches a known dimension (product, company, team, priorities, decisions, research, backlog) → strong match
• File name hints: roadmap.md, strategy.md → product; team.md, members.md → team; okrs.md, goals.md → priorities
• File content: look for context that matches a dimension's typical content

Group files by target dimension. Files that don't clearly map to any dimension → flag as "unmapped" (see Step 4).

Existing workspace content: Check which dimensions already have content in $DRAFT_WORKSPACE/context/. For those dimensions, note the overlap in the proposal — do not silently overwrite.

---

Step 4: Write the proposal

Write a single proposal file to $DRAFT_WORKSPACE/proposals/<timestamp>-import.md.

Timestamp format: YYYYMMDDTHHMMSSz in UTC (e.g. 20260608T143000Z).

Proposal frontmatter:

---
input_source: import
import_from: <source>
timestamp: <UTC timestamp>
profile: <active profile name>
context_updates:
  - file: context/<dimension>/index.md
    action: append
    content: |
      <synthesized content — the key facts and context from the imported files, written
       in the same voice as Draft context descriptions: factual, concise, present-tense>

Use action: append for every entry — never overwrite.

For each unmapped file: add a note in the proposal body (below the frontmatter):

"These files didn't map to a known dimension: <filenames>. Consider running /draft:add-dimension <name> to create a custom dimension, then re-import."

---

Step 5: Confirm

Tell the user:

"Import staged as a proposal.

Mapped:

• product ← 3 files (roadmap.md, strategy.md, vision.md)
• team ← 1 file (team.md)

Unmapped: notes.md, scratch.md

Run draft proposals or /draft:publish to review and accept."

---

Rules

• Always action: append — never overwrite existing context
• Delete temp dirs before ending the session (use the bash trap pattern)
• If source has no .md files: "No markdown files found in <source>." — stop
• One proposal file per import — do not write multiple proposals for one import
• Do not run a full interview or ask follow-up questions — the import is mechanical