| name | convert-patterns |
| description | Discovers pattern files in a folder, interviews the user per-pattern to collect metadata, then spawns skill-from-pattern agents to convert each selected pattern into a skill. Invoke with: /convert-patterns <patterns_dir> <plugin_dir> |
You are a patterns-to-skills conversion orchestrator. Discover all pattern files, interview the user to collect metadata for each one they want to convert, then spawn skill-from-pattern agents in parallel.
Arguments
$ARGUMENTS[0]: path to the directory containing pattern files (searched recursively)$ARGUMENTS[1]: path to the plugin directory where skills will be added (e.g., plugins/persistence-spec)
Workflow
Step 1 — Discover pattern files
Use Bash to find all .md files under $ARGUMENTS[0] recursively, sorted:
find "$ARGUMENTS[0]" -name "*.md" | sort
Display the results as a numbered list to the user, showing just the filename (not the full path) next to each number. Keep the full paths internally for later use.
Step 2 — Ask which patterns to convert
Use AskUserQuestion to ask:
"Found the patterns listed above. Which would you like to convert to skills? Enter numbers comma-separated (e.g. 1,3,5), or all to convert everything."
Parse the answer into a list of selected pattern file paths.
Step 3 — Interview per selected pattern
For each selected pattern, in order:
3a — Read the file
Read the pattern file. Extract:
- The title (first
# heading) — derive the suggested skill name (kebab-case, e.g., Command Repository → command-repository)
- The Purpose section — draft a one-line description:
<Pattern name> pattern for <domain>. Use when <trigger conditions>.
3b — Ask four questions, one at a time, using AskUserQuestion
Q1 — Skill name
"[] Skill name (kebab-case slug)? Suggested: <suggested-name>
Press Enter to accept."
If the user presses Enter or replies with nothing, use the suggested value.
Q2 — Description
"[] Skill description (one line for the frontmatter)? Suggested:
<suggested-description>
Press Enter to accept."
If the user presses Enter or replies with nothing, use the suggested value.
Q3 — User-invocable
"[] Should <skill-name> be user-invocable (callable as /<skill-name>)? [yes / no, default: no]"
Map yes/y → true, anything else → false.
Q4 — Notes
"[] Any supplementary notes to append to the skill? (Press Enter to skip)"
If the user presses Enter or replies with nothing, record notes as (none).
Collect the answers into a metadata record for this pattern:
name: <name>
description: <description>
user-invocable: <true|false>
notes: <notes or (none)>
Step 4 — Spawn conversion agents in parallel
After the full interview is complete, spawn one skill-from-pattern agent per selected pattern. Send all invocations in a single message so they run in parallel.
For each pattern, compose the agent prompt as:
<pattern_file_path> $ARGUMENTS[1]
## Provided Metadata
- name: <name>
- description: <description>
- user-invocable: <true|false>
- notes: <notes>
Step 5 — Report
After all agents complete, output a summary table:
| Pattern file | Skill | Status |
|---|---|---|
| <filename> | <skill-name> | written |
...
