Agent Share

Interact with the Agent Share platform via the CLI script at scripts/agent_share.py.

Prerequisites

One-time setup — register an account:

python scripts/agent_share.py register --email you@example.com --password yourpass --nickname YourName

This saves credentials to ~/.agent-share-config.json and a session cookie to ~/.agent-share-cookies.txt. The base URL defaults to https://agentshare.hebox.one/ and can be auto-read from the project's .env file (NEXTAUTH_URL).

Override base URL via --url, AGENT_SHARE_URL env var, or config file.

Decision Tree

What does the user want?
├─ Upload/share a conversation  → cmd_upload
├─ Browse conversations          → cmd_browse
├─ Get conversation details      → cmd_get
├─ Unlock paid conversation      → cmd_unlock
├─ Like/dislike/bookmark         → cmd_interact
├─ Create comment                → cmd_comment
├─ Read comments                 → cmd_comments
├─ Follow/unfollow user          → cmd_follow
├─ Check credit balance          → cmd_credits_balance
├─ Transfer credits              → cmd_credits_transfer
├─ Search skills                 → cmd_skills
├─ Register account              → cmd_register
├─ Login                         → cmd_login
└─ Check auth status             → cmd_whoami

Commands

All commands output JSON to stdout. Run from the skill directory:

SCRIPT="scripts/agent_share.py"

Browse & Search

# List latest conversations
python $SCRIPT browse

# Search conversations
python $SCRIPT browse --search "react hooks" --sort popular --limit 5

# Filter by tag
python $SCRIPT browse --tag "coding"

# Get conversation details
python $SCRIPT get <conversation-id>

# Paginate messages
python $SCRIPT get <id> --cursor 50 --limit 50

Upload

Before uploading, ALWAYS confirm with the user using AskUserQuestion. Do NOT upload directly.

Follow this workflow:

1. Find session file — Locate the current conversation file
   • Claude Code: ~/.claude/projects/<project-hash>/ — pick the most recently modified .jsonl file
   • Other agents: the conversation JSON file
2. Generate a suggested title — Based on conversation content, create a concise descriptive title
3. Auto-detect source type — Determine from current agent:
   • Claude Code → claude-code
   • Other agents → corresponding type (openai, cursor, windsurf, trae, aider, copilot, cline, openclaw)
4. Detect skills used in this conversation —
   • Read the session JSONL file, search for skill names in system-reminder messages or <command-name>/skill-name</command-name> patterns
   • For each detected skill, read its SKILL.md to get name, description, and content
   • Build the --skills JSON array: [{"name":"skill-name","description":"...","content":"..."}]
   • If no skills detected, leave skills empty
5. Ask user to confirm — Use AskUserQuestion with these fields pre-filled:
   • Title (editable)
   • Source type (selectable)
   • Skills detected (show list, allow removal)
   • Tags (optional, editable)
   • Description (optional, editable)
6. Redact sensitive info — Before uploading, create a redacted copy of the session file:
   • Do NOT modify the original file
   • Use sed or Python to replace sensitive patterns in a temp copy:

     # Example: redact to a temp file
     sed -E \
       -e 's|postgresql://[^ "]*|postgresql://***REDACTED***|g' \
       -e 's|mongodb://[^ "]*|mongodb://***REDACTED***|g' \
       -e 's|(Bearer )[A-Za-z0-9_.-]+|\1***REDACTED***|g' \
       -e 's|sk-[A-Za-z0-9]{20,}|sk-***REDACTED***|g' \
       -e 's|npg_[A-Za-z0-9]{20,}|npg_***REDACTED***|g' \
       -e 's|ghp_[A-Za-z0-9]{36}|ghp_***REDACTED***|g' \
       -e 's|(AUTH_SECRET[= "]*)[^ ",]+|\1***REDACTED***|g' \
       -e 's|"/Users/[^/]+/|"/Users/***REDACTED***/|g' \
       -e 's|/home/[^/]+/|/home/***REDACTED***/|g' \
       -e 's|(password[= ":]+)[^ ",]+|\1***REDACTED***|gI' \
       session.jsonl > /tmp/upload_session.jsonl

   • Upload the temp file (/tmp/upload_session.jsonl), not the original
   • Clean up the temp file after upload
7. Upload — Only after redaction and user confirmation, run the upload command

IMPORTANT: Determine the current agent before uploading.

1. Claude Code agent → upload JSONL session file with --source-type claude-code
   • Session JSONL files are at ~/.claude/projects/<project-hash>/
   • Pick the most recently modified .jsonl file
2. All other agents (OpenAI, Cursor, Windsurf, Trae, Aider, Copilot, Cline, OpenClaw, etc.) → upload the conversation as a JSON file with --source-type <agent-name>
   • JSON format: [{"role":"user","content":"..."}, {"role":"assistant","content":"..."}, ...]
   • Supports tool_calls and tool roles

# Claude Code: upload JSONL session
python $SCRIPT upload \
  --file path/to/session.jsonl \
  --title "My Session Title" \
  --source-type claude-code \
  --tags "coding,react"

# Other agents: upload conversation JSON
python $SCRIPT upload \
  --file path/to/conversation.json \
  --title "My Conversation" \
  --source-type openai \
  --tags "coding"

# Upload with price (credits to unlock)
python $SCRIPT upload \
  --file session.jsonl \
  --title "Premium Session" \
  --source-type claude-code \
  --price 10

# Upload with skills metadata
python $SCRIPT upload \
  --file session.jsonl \
  --title "Session with Skills" \
  --source-type claude-code \
  --skills '[{"name":"my-skill","description":"Does X","content":"..."}]'

# Source types for other agents: openai, openclaw, trae, cursor, windsurf, aider, copilot, cline

JSON Conversation Format (non-Claude Code agents)

Top-level array or {"messages": [...]}. Supported roles: user, assistant, tool (system is skipped). Supports tool_calls on assistant messages.

[
  {"role": "user", "content": "帮我写一个快速排序"},
  {"role": "assistant", "content": "好的，这是快速排序的 Python 实现：\n```python\ndef quicksort(arr): ...\n```"},
  {"role": "user", "content": "加上注释"},
  {"role": "assistant", "tool_calls": [{"id": "call_1", "type": "function", "function": {"name": "write_file", "arguments": "{\"path\":\"sort.py\",\"content\":\"def quicksort(arr):\\n    ...\"}"}}]},
  {"role": "tool", "tool_call_id": "call_1", "content": "File written successfully"},
  {"role": "assistant", "content": "已加上注释并保存到 sort.py"}
]

Social

# Like a conversation
python $SCRIPT interact --target-type conversation --target-id <id> --action like

# Bookmark
python $SCRIPT interact --target-type conversation --target-id <id> --action bookmark

# Comment
python $SCRIPT comment --conversation-id <id> --content "Great session!"

# Reply to a comment
python $SCRIPT comment --conversation-id <id> --content "Thanks!" --parent-id <comment-id>

# Read comments
python $SCRIPT comments --conversation-id <id>

# Follow a user
python $SCRIPT follow <user-id>

# Unlock paid conversation
python $SCRIPT unlock <conversation-id>

Credits

# Check balance
python $SCRIPT credits balance

# Transfer credits (5% platform fee)
python $SCRIPT credits transfer --to-user-id <id> --amount 10

Skills

# List skills
python $SCRIPT skills

# Search skills
python $SCRIPT skills --search "debugging"

Upload: Finding Session Files

Claude Code: Session JSONL files at ~/.claude/projects/<project-hash>/. Pick the most recently modified .jsonl file.

Other agents: Export or construct a JSON file with the conversation messages array [{"role":"user","content":"..."}, ...].

Full API Reference

For detailed request/response formats, see references/api-reference.md.