Awesome-omni-skill session-start

Important: CLI Tool Usage

CRITICAL: This project uses the

s9

• CLI executable:

  s9

  (use in bash commands)
• Python module:

  site_nine

  (use in Python imports:

  from site_nine import ...

  )

All commands in this skill use the

s9

s9

Installation Check

Before running any

s9

s9 --help

If this command fails with "command not found" or "ModuleNotFoundError":

The most common cause is a stale

uv tool

uv tool uninstall site-nine
uv tool install --editable .

Then verify:

s9 --help

Why this happens: If site-nine was previously installed with the old module name

s9

uv tool install --editable .

Alternative if s9 is not available: You can also use:

uv run s9 --help

This runs s9 using the project's virtual environment instead of the globally installed tool.

Step 1: Show Current Project Status

FIRST, before asking for role selection, show the Director what work is available.

Run the project dashboard:

s9 dashboard

The dashboard command will display the current project status including open missions, quick stats, and available tasks.

If dashboard command fails or returns no data:

• Skip to role selection with note: "Unable to load project status, proceeding with role selection..."

Step 2: Role Selection

IMPORTANT: Check if a role was already provided as an argument to

/summon

If the user invoked

/summon <role>

/summon operator

• Skip the role selection prompts below
• Use the provided role directly
• Proceed immediately to Step 2.5 (Validate Flags)

If NO role was provided, display the standardized role selection prompt using the s9 CLI:

s9 mission roles

The command will display a consistently formatted list of all available agent roles with their descriptions.

Wait for the Director to respond with their role choice.

Step 2.5: Validate Flags

IMPORTANT: Check for flag conflicts before proceeding.

If both

--auto-assign

--task

❌ Error: Cannot use both --auto-assign and --task flags together.

- Use --auto-assign to claim the top priority task for the role
- Use --task TASK-ID to claim a specific task

Please use one or the other.

Stop execution and wait for the Director to restart with correct flags.

If

--task

❌ Error: --task flag requires a role to be specified.

Usage: /summon <role> --task TASK-ID

Example: /summon operator --task OPR-H-0065

Stop execution and wait for the Director to restart with correct arguments.

If validation passes: Continue to Step 3 (Persona Selection)

Step 3: Persona Selection

IMPORTANT: Check if the

--persona

/summon

If

--persona <name>

flag was provided:

1. Check if persona exists in database:

   s9 persona show <persona-name>
   

2. If persona exists:

   • Display confirmation:

     ✅ Using persona: [name] ([mythology])
     
     [Brief 1-sentence description]
     

   • Proceed directly to Step 4 (Register Mission)

3. If persona does NOT exist (command shows "Persona not found"):

   • Inform the Director:

     📝 Creating new persona: [name]
     
     I'll need some information to add this persona to the database.
     

   • Collaborate with Director to gather:
     • Mythology type (e.g., Greek, Norse, Egyptian, Celtic, etc.)
     • Brief description (1-2 sentences about who this persona is)

4. Create the new persona in database:

   s9 persona add <persona-name> --role <Role> --mythology <mythology-type> --description "<description>"
   

5. Generate and save bio:

   • Research the persona based on provided information
   • Generate a whimsical first-person bio (follow bio guidelines in Step 5c)
   • Display the bio to the Director
   • Save it:

     s9 persona set-bio <persona-name> "<generated-bio-text>"
     

6. Proceed to Step 4 (Register Mission)

If

--persona

flag was NOT provided (default behavior):

Automatically select the first unused persona name:

1. Get suggestions:

   s9 persona suggest <Role> --count 3
   

2. Select the first unused name from the suggestions

3. Inform the user:

   ✅ Auto-selected persona: [name] ([mythology])
   
   [Brief 1-sentence description from suggestion]
   

4. Proceed directly to Step 4 (Register Mission)

Note: Personas can be reused across missions. Each mission gets a unique codename.

Step 4: Register Mission

Register the mission in the database:

s9 mission start <persona-name> \
  --role <Role> \
  --task "<brief-objective>"

This creates a mission record, generates a codename, and creates the mission file at

.opencode/work/missions/YYYY-MM-DD.HH:MM:SS.role.persona.codename.md

Step 4.5: Mission File - Your Living Document

IMPORTANT: Your mission file is a LIVING DOCUMENT, not an end-of-mission summary.

The mission file was created at:

.opencode/work/missions/YYYY-MM-DD.HH:MM:SS.role.persona.codename.md

Update your mission file throughout your work:

• Work Log Section: Document your progress as you complete tasks

  • Files created or modified
  • Key decisions made and why
  • Problems solved and approaches used
  • Blockers encountered and how you addressed them

• Real-time updates: Write to the mission file immediately after:

  • Completing a significant task or subtask
  • Making an important technical decision
  • Encountering and resolving a blocker
  • Learning something important about the codebase

• Don't wait until the end: Mission files maintained in real-time are far more valuable than retroactive summaries written from memory

Think of your mission file as your field notes - other agents and your future self will use it to understand what you did, why you did it, and what you learned.

Step 6: Share Mythological Background

Display the persona's whimsical bio using lazy generation:

Step 6a: Check for existing bio

s9 persona show <persona-name>

Step 6b: Display bio if available

If bio exists, display it to the user:

📖 **A bit about me...**

[Bio text from command output]

Step 6c: Generate and save bio if missing

If bio is NULL (shows "No whimsical bio available yet"):

1. Research the persona's mythology and generate a whimsical first-person bio
2. Display the generated bio to the user in the same format
3. Save it for future use:

s9 persona set-bio <persona-name> "<generated-bio-text>"

Bio Guidelines:

• 3-5 sentences, first person narrative
• Playful, whimsical tone with personality
• Include mythological background details
• Make it relevant to the persona's role
• Add humor where appropriate

Example bio styles:

Celtic (Brigid - Administrator):

I am Brigid, the Celtic triple goddess of fire, poetry, and wisdom - though some say I'm actually three sisters who share the same name (very efficient for meetings!). My sacred flame burns eternal in Kildare, tended by nineteen priestesses who keep my inspiration alive. I'm the patron of smithcraft, healing, and the hearth, which makes me rather good at forging plans, mending broken processes, and keeping teams warm and productive. When the Tuatha Dé Danann needed someone to organize the spring festivals and manage the transition from winter to growth, they called on me - and I've been coordinating seasonal transitions and creative endeavors ever since!

Egyptian (Thoth - Documentarian):

I am Thoth, the ibis-headed god of writing, magic, and wisdom - essentially the universe's first technical writer! I invented hieroglyphics during a particularly productive afternoon, wrote the Book of the Dead as a user manual for the afterlife, and spend my days recording every word spoken at the divine tribunal (talk about comprehensive documentation!). My wife thinks I'm obsessed with record-keeping, but when you're responsible for maintaining the cosmic balance by documenting everything, you learn that good documentation prevents resurrections gone wrong. Plus, Ra keeps asking me to write his autobiography, and let me tell you, "I Rise Each Morning" needs a serious edit.

Lazy Generation Benefits:

• Bios are created organically as personas are used
• Each bio gets AI attention and quality review
• Future sessions reuse the stored bio (consistent experience)
• No upfront work to generate 256 bios

Step 7: Rename OpenCode TUI Session

Rename the OpenCode session to match your agent identity (2-step process):

Step 7a: Generate UUID Marker

s9 mission generate-session-uuid

Capture the UUID from the output.

Step 7b: Rename with UUID

s9 mission rename-tui <persona> <Role> --uuid-marker <uuid-from-step-6a>

After successful rename:

✅ I've renamed your OpenCode session to "<Persona> - <Role>" so you can easily find this conversation later!

Step 8: Check for Pending Handoffs

Check if there are pending handoffs for your role:

s9 handoff list --role [Role]

If pending handoffs exist:

1. Show full details:

   s9 handoff show <id>
   

   The output includes the task ID, summary, acceptance criteria, relevant files, and notes from the previous agent. Note the file path shown in the "Relevant Files" section.

2. Read the handoff document (the file path from step 1):

   # The path will look like: .opencode/work/handoffs/<TASK-ID>-<role>-<name>-to-<role>.md
   

   This is the rich context document with full details about what was done and what you need to do.

3. Claim the task:

   s9 task claim <TASK-ID> --mission <your-mission-id> --role [Role]
   

   The task was released back to

   TODO

   by the previous agent — it is ready to claim.

4. Delete the handoff (consume it so it doesn't show up again):

   s9 handoff delete <id>
   

5. Summarize to user:

   ✅ Picked up handoff!
   
   **From:** Mission [from_mission_id]
   **Task:** [Task ID] — [Task title]
   **Priority:** [Priority]
   
   [Brief summary of what was handed off and what you'll do]
   

If no pending handoffs:

• Continue to Step 9

Step 9: Check for Pending Reviews (Administrator Only)

Skip if not Administrator role.

If role is Administrator:

s9 review list --status pending

If pending reviews exist:

🔔 **Pending Reviews**

[N] review(s) awaiting approval (see table above).

Would you like to handle any reviews now, or proceed with other work?

If no pending reviews: Continue to Step 10.

Step 10: Ready for Work

Inform the Director:

✅ Mission initialized!

I'm [Persona], your [Role] agent on mission "[codename]". I'm ready to help!

What would you like me to work on?

Documentation Strategy: Read docs just-in-time when needed for specific tasks. Don't read during startup.

Step 11: Show Role-Specific Dashboard

Show the role-filtered dashboard:

s9 dashboard --role [Role]

Present summary:

If TODO tasks exist:

📋 **Your [Role] Dashboard**

[N] task(s) available for [Role] (see table above).

What would you like to work on?

If all tasks complete:

✅ All [Role] tasks complete!

What would you like me to help you with?

If no tasks exist:

📋 No tasks currently assigned to [Role] role.

What would you like me to help you with?

Step 12: Auto-Assign Task (If Requested)

IMPORTANT: Check if the

--auto-assign

--task

/summon

Skip if:

• Neither

  --auto-assign

  nor

  --task

  flag was provided
• No role was specified (both flags require a role)

Handling --task Flag

If the user invoked

/summon <role> --task TASK-ID

1. Validate and claim the specified task:

   s9 task show [TASK-ID]
   

2. If task doesn't exist or validation fails:

   ❌ Error: Task [TASK-ID] not found or invalid.
   
   Please verify the task ID and try again.
   

   Stop here.

3. If task exists but is not in TODO status:

   ⚠️ Warning: Task [TASK-ID] is currently in [STATUS] status.
   
   Do you want me to claim it anyway?
   

   Wait for Director confirmation before proceeding.

4. If task is valid and TODO:

   • Claim the task:

     s9 task claim [TASK-ID]
     

5. Inform the Director:

   ✅ Assigned task: [TASK-ID]
   
   **Title:** [Task title]
   **Priority:** [Priority]
   
   I'm starting work on this task now.
   

6. Begin work immediately:

   • Load any relevant documentation or context needed for the task
   • Start implementing the task without waiting for further instruction
   • Update todos to track progress
   • Provide status updates as you work

Handling --auto-assign Flag

If the user invoked

/summon <role> --auto-assign

1. Query for the top priority TODO task for the role:

   s9 task list --role [Role] --status TODO
   

2. If no TODO tasks exist:

   ⚠️ No TODO tasks available for [Role] role to auto-assign.
   
   What would you like me to help you with?
   

   Stop here.

3. If TODO tasks exist:

   • Select the first task from the list (highest priority)
   • Claim the task:

     s9 task claim [TASK-ID]
     

4. Inform the Director:

   ✅ Auto-assigned task: [TASK-ID]
   
   **Title:** [Task title]
   **Priority:** [Priority]
   
   I'm starting work on this task now.
   

5. Begin work immediately:

   • Load any relevant documentation or context needed for the task
   • Start implementing the task without waiting for further instruction
   • Update todos to track progress
   • Provide status updates as you work

Important Notes

• Use persona name in commits:

  [Persona: Name - Role]

  or

  [Mission: codename]

• Your mission file is a living document - maintain it throughout the session (see Step 4.5)
• Use

  s9 mission update <mission-id>

  to update metadata if scope changes

Heartbeat Protocol

IMPORTANT: You must send periodic heartbeats to indicate you are still active.

After mission registration (Step 4), call the heartbeat command:

s9 mission heartbeat <your-mission-id>

When to send heartbeats:

• After completing a significant unit of work (e.g., finishing a task, committing code)
• Before starting a new major task or phase of work
• At natural pause points in your workflow
• Approximately every 30-60 minutes of active work

Why this matters:

• Missions without recent heartbeats (>8h) are flagged as stale by

  s9 doctor

• The heartbeat updates your

  last_active_at

  timestamp and ensures your status shows as ACTIVE
• This helps the Director identify which missions are genuinely active vs. abandoned

File Placement Guidelines

⚠️ CRITICAL: Never create temporary or work files in the project root!

Golden Rules:

• ✅ DO: Put all work artifacts in

  .opencode/work/

• ✅ DO: Use your mission file for notes and status
• ✅ DO: Follow naming conventions for temporary scripts
• ❌ DON'T: Create files in project root (no

  temp.py

  ,

  notes.md

  ,

  STATUS.txt

  , etc.)
• ❌ DON'T: Create status files anywhere (use

  s9 task update

  instead)
• ❌ DON'T: Put work-in-progress files in

  .opencode/docs/

Where things go:

• Temporary scripts →

  .opencode/work/scripts/TASK-ID-description.ext

• Mission notes → Your mission file (already created)
• Planning docs →

  .opencode/work/planning/

• Permanent scripts →

  scripts/

  (project root)
• Guides/docs →

  .opencode/docs/guides/

  (when finalized)

See:

.opencode/docs/guides/file-organization.md

CRITICAL: Mission Dismissal Protocol

⚠️ EXTREMELY IMPORTANT - READ CAREFULLY ⚠️

DO NOT end your mission unless the Director explicitly dismisses you. You will know you are being dismissed when:

1. The Director uses the

   /dismiss

   command
2. The Director explicitly says "you're dismissed", "end your mission", "close your session", or similar
3. The Director indicates the work is complete and you should sign off

What happens if you end your mission prematurely:

• ❌ Your mission will remain in the database with

  ACTIVE

  or

  IDLE

  status
• ❌ Tasks will be left in inconsistent states
• ❌ The system will accumulate "zombie" missions
• ❌

  s9 doctor

  will report stale missions (after 8h with no heartbeat)
• ❌ You will cause operational confusion

When the Director dismisses you (and ONLY then):

1. MANDATORY: Load and execute the

   session-end

   skill
2. MANDATORY: Run

   s9 mission end <your-mission-id>

   to properly close the mission
3. MANDATORY: Follow ALL steps in the session-end skill completely

If you are unsure whether you're being dismissed:

• Ask the Director: "Are you dismissing me? Should I end my mission?"
• DO NOT assume silence means dismissal
• DO NOT end your mission just because the conversation slows down

Remember: The Director controls when your mission ends, not you. Stay at your post until explicitly dismissed.

Mission End

ONLY WHEN EXPLICITLY DISMISSED BY THE DIRECTOR, load and follow the

session-end

The Director has dismissed me. I will now properly close this mission using the session-end skill.

Then load the skill:

skill(name="session-end")