| name | agentic-share |
| description | Shared core logic for importing/exporting reusable assets into the v0.2 plugin architecture. Internal helper for agentic-import and agentic-export. |
| project-agnostic | false |
| allowed-tools | ["Read","Write","Edit","Bash","Glob","Grep"] |
Agentic Asset Share
Core logic for moving reusable assets between projects and this repository.
Arguments
- mode:
import | export
- asset_type:
skill | template | agent
- source: source path or asset name
- target: optional destination name/path
- options: optional flags (
--plugin, --force, --dry-run)
Parsed from: $ARGUMENTS
Canonical Locations (v0.2)
| Type | Project Source | Repository Target |
|---|
| skill | .claude/skills/<name>/ or explicit skill dir | plugins/<plugin>/skills/<name>/ |
| template | templates/<name>/ | templates/<name>/ |
| agent | explicit agent markdown path | plugins/ac-workflow/agents/<name>.md |
Notes:
- There is no command asset type in v0.2 (skills-only surface).
- For skills,
--plugin is required unless destination can be inferred.
Execution Flow
- Parse mode, type, source, target, options.
- Validate mode (
import or export).
- Validate asset type (
skill, template, agent).
- Validate repository root:
- must contain
.claude-plugin/marketplace.json
- must contain
plugins/
- Resolve source and destination paths by type.
- Read source content (single file or directory tree).
- Apply sanitization:
- scrub absolute home paths (
/Users/..., /home/...) -> {USER_HOME}
- scrub secrets/tokens (
api_key, token, secret, bearer values) -> {REDACTED}
- Validate sanitized output:
- block if unresolved secrets remain
- warn on environment-specific hardcoded values
- Write output (or preview with
--dry-run).
- Report summary and next steps.
Safety Rules
- Never overwrite existing targets without
--force or explicit confirmation.
- Never proceed when unresolved secrets remain after sanitization.
- On any failure, abort with a clear error and leave no partial file writes.
Commit Guidance (Optional)
After successful import/export, offer a commit for only touched asset paths.
Conventional message pattern:
feat(assets): import <asset_type> '<asset_name>'
(or export accordingly)
Error Cases
| Error | Action |
|---|
| Invalid mode/type | Abort with valid choices |
| Repository root not detected | Abort and request running from repo root |
| Source missing | Abort and print source path |
| Sanitization blockers remain | Abort and print blocker lines |
| Target exists without force | Ask for overwrite confirmation |
