manus-skill（手艺）

Manus API Skill

Manage tasks, projects, files, and sessions on the Manus AI agent platform.

Setup

export MANUS_API_KEY="your-api-key"

Get your key from: Manus dashboard → Settings → Integration → Build with Manus API.

The session manager writes .manus_sessions.json to the project root to track active and archived sessions. Add it to .gitignore — it contains prompts, task IDs, and task URLs that may be sensitive.

echo ".manus_sessions.json" >> .gitignore

Authentication

Every request uses the API_KEY header (not Bearer token):

curl -H "API_KEY: $MANUS_API_KEY" https://api.manus.ai/v1/tasks

Base URL: https://api.manus.ai — all endpoints under /v1/.

Session Management (CRITICAL — Read First)

Every POST /v1/tasks without a taskId creates a brand-new task and consumes credits. There is no undo. Session tracking prevents accidental duplication.

For the full registry format, search algorithm, archival rules, and edge cases, read references/session-management.md.

How It Works

All sessions are tracked in .manus_sessions.json in the project root:

{
  "active": {
    "security-audit": { "task_id": "...", "description": "...", "tags": [...], ... }
  },
  "archived": {
    "q2-report": { "task_id": "...", "description": "...", "archived_reason": "completed", ... }
  }
}

Active = running or pending tasks. Archived = completed or failed (still searchable, can be resumed).

The 4 Rules

1. Before creating any task, search the registry for a matching session. If found, continue it with taskId. If not, create new.
2. After every task creation, save task_id + metadata to the registry immediately.
3. When multiple sessions match, present the top matches and ask the user to pick. Never guess.
4. When a task completes or fails, move it from active to archived automatically.

Session Resolution Flow

When the user wants to work with a session:

1. Search active sessions (name, description, tags, prompts)
      |
   1 match  -> Use it
   0 matches -> Search archived
   >1 match  -> Present list, ask user
      |
2. Search archived
      |
   1 match  -> Ask: "Archived (completed). Resume or start fresh?"
   0 matches -> Create new session
   >1 match  -> Present list, ask user

Registry Entry Fields

Each session stores: task_id, task_title, task_url, description (agent-generated summary), tags (keywords for search), project_id, project_name, agent_profile, task_mode, created_at, last_used, last_status, turn_count, first_prompt, last_prompt.

Archived entries additionally store archived_at and archived_reason.

Using the Session Manager Script

The script at scripts/manus_session.py handles all session tracking, API calls, polling, archival, and search. Use it instead of calling the API directly.

As a library (when an agent imports it):

from scripts.manus_session import ManusSession

s = ManusSession()  # reads .env automatically

# New session
s.send("Review auth module for vulnerabilities",
       session_name="security-audit",
       tags=["security", "auth", "code-review"])

# Continue (auto-passes taskId)
s.send("Now check the token refresh logic",
       session_name="security-audit")

# Search sessions
matches = s.search("auth")

# Poll until done (auto-archives on completion)
result = s.poll_until_done("security-audit")

# List sessions
s.list_sessions()                        # active only
s.list_sessions(include_archived=True)   # both

# Resume archived session
s.unarchive("q2-report")
s.send("Add December data", session_name="q2-report")

# Import orphaned task from Manus webapp
s.import_task("TeBim6FDQf9peS52xHtAyh", session_name="imported-task")

As a CLI (when an agent runs it from the shell):

# Send prompt (new or continue)
python scripts/manus_session.py send "Review this PR" -s pr-review --tags security,backend
python scripts/manus_session.py send "Focus on auth" -s pr-review  # continues

# Search
python scripts/manus_session.py search "auth"

# List sessions
python scripts/manus_session.py sessions
python scripts/manus_session.py sessions --archived

# Check status
python scripts/manus_session.py status -s pr-review

# Poll until done
python scripts/manus_session.py poll -s pr-review

# Resume archived
python scripts/manus_session.py unarchive -s old-session

# Import orphaned task
python scripts/manus_session.py import-task TASK_ID --name my-task

# Upload file
python scripts/manus_session.py upload report.pdf

# Projects
python scripts/manus_session.py create-project "Code Reviews" --instruction "Check for security issues"
python scripts/manus_session.py projects

# Cleanup old archived sessions (>7 days)
python scripts/manus_session.py cleanup

Quick Reference — Endpoints

For full request/response schemas, read references/api-reference.md.

Core Workflows

1. Create a New Session

python scripts/manus_session.py send \
  "Analyze Q2 revenue trends and create a summary report" \
  -s q2-analysis --mode agent --tags revenue,q2,analysis

2. Continue an Existing Session

python scripts/manus_session.py send \
  "Now break it down by region" -s q2-analysis

3. Upload File and Attach

python scripts/manus_session.py upload report.pdf
# Returns file_id

python scripts/manus_session.py send \
  "Summarize this document" -s doc-review --file-id FILE_ID

Files expire after 48 hours.

4. Organize with Projects

python scripts/manus_session.py create-project "Research" \
  --instruction "Always cite sources and include confidence levels"

python scripts/manus_session.py send \
  "Research AI agent frameworks" -s agent-research --project proj_abc123

Agent Profiles & Task Modes

Connectors

Enable external service access by passing connector UUIDs:

• Gmail: 9444d960-ab7e-450f-9cb9-b9467fb0adda
• Notion / Google Calendar: configured per account

Cost Awareness

• ~150 credits per typical task. Use manus-1.6-lite + chat for simple queries.
• Always reuse sessions via taskId. Duplicate tasks = wasted credits.
• Check credit_usage in task responses to monitor spend.

Reference Files