| name | claude-skill-reclassify-agents |
| description | Reclassify existing Claude Code subagents and skills, then migrate files into the appropriate structure with backups and a report. |
Skill: Reclassify existing subagents vs skills and migrate
Intent
Scan an existing repo for Claude Code subagents and skills, decide which each item should be, and apply the conversion by moving/rewrapping files into the appropriate structure — with backups and a migration report.
This skill is designed for repos that contain:
- subagents as Markdown files (commonly
.claude/agents/*.md)
- skills as either:
- folder-based skills (
.claude/skills/<skill>/SKILL.md), or
- single Markdown skills (various conventions)
If your structure differs, adjust config.yaml.
Inputs you need from the user (or derive automatically)
- repository root (assume current working directory is repo root)
- optional: path to config (
config.yaml)
- desired mode:
dry-run (generate report only)apply (convert + update references)
Outputs
reclassify-report.md
- list of items
- current type (agent/skill/unknown)
- recommended type
- confidence score
- top rationale signals
- migration plan summary
- backups under
.reclassify_backup/<timestamp>/...
- migrated files in:
target_agents_dir (default .claude/agents)target_skills_dir (default .claude/skills/<name>/SKILL.md)
Operating rules (guardrails)
- Always create a backup snapshot under
.reclassify_backup/... before writing.
- Prefer running on a git branch (create one if possible).
- Never delete originals; if moving, keep an embedded appendix OR keep an original copy under backup.
- If an item references tools/paths that do not exist, preserve the text; do not “fix” behavior beyond migration.
- Update references best-effort (rename mentions), but do not break code; keep replacements conservative.
Procedure (high-level)
- Discover existing agents & skills by scanning configured directories (and auto-detect when missing).
- Classify each item using the rubric:
- Skill score ↑ with procedural/template/checklist/format markers
- Subagent score ↑ with role/principles/judgment/exploration markers
- Report: write
reclassify-report.md with confidence & rationale.
- Convert:
- Agent → Skill:
- create folder
target_skills_dir/<name>/SKILL.md
- rewrap content into a Skill template (Goal / Inputs / Outputs / Procedure / Guardrails)
- Skill → Agent:
- create
target_agents_dir/<name>.md
- rewrap content into an Agent template (Role / Principles / How to respond)
- Reference updates (optional):
- search Markdown files and replace references to moved paths/names where safe.
- Summary: print a small summary and point to report + backup path.
How to run (recommended)
Use the provided script to execute the procedure end-to-end:
python3 .claude/skills/reclassify-agents/scripts/reclassify.py --config .claude/skills/reclassify-agents/config.yaml --mode dry-run
python3 .claude/skills/reclassify-agents/scripts/reclassify.py --config .claude/skills/reclassify-agents/config.yaml --mode apply
