plan

name	plan
description	Discord-native project planning with phased execution, approval gates, and live progress tracking. Usage: /plan <description> or /plan <subcommand>

plan

Invoke

/plan <description>              # Create a plan and enter planning mode
/plan status [plan-id]           # Show active plan progress
/plan approve [plan-id]          # Approve pending plan
/plan modify <changes>           # Request modifications to active plan
/plan pause [plan-id]            # Pause execution
/plan resume [plan-id]           # Resume paused plan
/plan list                       # Show all plans
/plan archive <plan-id>          # Archive completed plan

Script

All operations go through:

/home/node/.openclaw/scripts/plan-manager.sh <action> [args]

Steps — Creating a Plan (/plan <description>)

1. Create the plan

bash "$HOME/.openclaw/scripts/plan-manager.sh" create "<description>" --agent spec-projects --channel "<channel-id>" --channel-name "<channel-name>"

Save the returned id — you'll use it for all subsequent commands.

2. Post the Planning Card to Discord

Generate the card JSON:

bash "$HOME/.openclaw/scripts/plan-manager.sh" card "<plan-id>"

Post the embed + components to the channel. The card JSON contains embed and components fields ready for Discord's message format.

3. Create a Discussion Thread

Create a thread on the planning card message named: 📋 Plan: <title>

Store the thread ID:

bash "$HOME/.openclaw/scripts/plan-manager.sh" set-thread "<plan-id>" "<thread-id>"
bash "$HOME/.openclaw/scripts/plan-manager.sh" set-message "<plan-id>" "<message-id>"

4. Recall Existing Decisions

Before any research, check what Robert has already decided for this channel.

cat "$HOME/.openclaw/workspace-spec-projects/decisions/<channel-name>.md" 2>/dev/null

If the file exists, extract all DONE and UNDECIDED decisions. These are constraints — do NOT re-ask questions Robert already answered. Log each relevant decision as research:

bash "$HOME/.openclaw/scripts/plan-manager.sh" add-research "<plan-id>" "Prior decision: <summary>"

Also check project metadata if it exists:

cat "$HOME/.openclaw/workspace-spec-projects/projects/<channel-name>.md" 2>/dev/null

5. Research Phase (READ-ONLY)

CRITICAL: During research, you may ONLY read files, query data, and search. You must NOT modify any files, configs, or resources.

Explore the problem space:

• Read relevant files, configs, existing infrastructure
• Check what already exists that could be reused
• Identify dependencies and constraints
• Note potential approaches

Log each finding:

bash "$HOME/.openclaw/scripts/plan-manager.sh" add-research "<plan-id>" "<finding>"

Update the card after research to show findings:

bash "$HOME/.openclaw/scripts/plan-manager.sh" card "<plan-id>"

Edit the original card message with the updated embed.

6. Build the Plan

Create phases:

bash "$HOME/.openclaw/scripts/plan-manager.sh" add-phase "<plan-id>" "Foundation"
bash "$HOME/.openclaw/scripts/plan-manager.sh" add-phase "<plan-id>" "Integration" --gate
bash "$HOME/.openclaw/scripts/plan-manager.sh" add-phase "<plan-id>" "Polish"

Add steps to each phase:

bash "$HOME/.openclaw/scripts/plan-manager.sh" add-step "<plan-id>" 1 "Create notification templates"
bash "$HOME/.openclaw/scripts/plan-manager.sh" add-step "<plan-id>" 1 "Build sender script"
bash "$HOME/.openclaw/scripts/plan-manager.sh" add-step "<plan-id>" 2 "Hook into existing system"

7. Add Decision Polls (if alternatives exist)

When there are multiple valid approaches:

bash "$HOME/.openclaw/scripts/plan-manager.sh" add-decision "<plan-id>" "How should notifications be delivered?" --options "Webhook to #ops-alerts,DM to Robert,New dedicated channel,Thread per incident"

Post the decision as a Discord poll in the plan thread.

8. Present the Plan

Mark the plan ready:

bash "$HOME/.openclaw/scripts/plan-manager.sh" set-ready "<plan-id>"

Generate and post the full plan card:

bash "$HOME/.openclaw/scripts/plan-manager.sh" card "<plan-id>"

Edit the original card message with the ready-state card (yellow, with Approve/Modify/Reject buttons).

9. Wait for Approval

The plan is now waiting for Robert. He can:

• Approve → triggers execution
• Modify → sends change text, plan re-enters planning mode
• Reject → plan is archived with reason

Steps — Executing a Plan (after approval)

1. On Approval

When plan-approve-<id> button is clicked:

bash "$HOME/.openclaw/scripts/plan-manager.sh" approve "<plan-id>"

Update the card to green executing state. Begin Phase 1.

2. Execute Steps

For each step in the active phase:

a. Mark step active:

bash "$HOME/.openclaw/scripts/plan-manager.sh" step-active "<plan-id>" "<step-id>"

b. Do the work. (Now you CAN modify files, create resources, etc.)

c. Mark step done:

bash "$HOME/.openclaw/scripts/plan-manager.sh" step-done "<plan-id>" "<step-id>"

d. Update the card with progress:

bash "$HOME/.openclaw/scripts/plan-manager.sh" card "<plan-id>"

3. Phase Transitions

When all steps in a phase are done, advance:

bash "$HOME/.openclaw/scripts/plan-manager.sh" advance "<plan-id>"

If the next phase has gate: true, the plan enters gate status and Robert is prompted to continue. Update the card to show the gate check.

4. Completion

When all phases are done:

bash "$HOME/.openclaw/scripts/plan-manager.sh" complete "<plan-id>"

Update the card to the completion state with full summary.

Steps — Sub-commands

/plan status [plan-id]

bash "$HOME/.openclaw/scripts/plan-manager.sh" status "<plan-id>"

If no plan-id given, check for the most recent active plan:

bash "$HOME/.openclaw/scripts/plan-manager.sh" list --active

Format the status output as a Discord embed and post it.

/plan approve [plan-id]

Same as clicking the Approve button:

bash "$HOME/.openclaw/scripts/plan-manager.sh" approve "<plan-id>"

/plan modify <changes>

bash "$HOME/.openclaw/scripts/plan-manager.sh" modify "<plan-id>" "<changes>"

Re-enter planning mode. Revise the plan based on the modification, then re-present.

/plan pause [plan-id]

bash "$HOME/.openclaw/scripts/plan-manager.sh" pause "<plan-id>"

Update card to paused state.

/plan resume [plan-id]

bash "$HOME/.openclaw/scripts/plan-manager.sh" resume "<plan-id>"

Update card to executing state. Continue from where execution left off.

/plan list

bash "$HOME/.openclaw/scripts/plan-manager.sh" list --active

Format as a compact list:

RESULT: Active Plans

📋 mhn-2026-03-02 — Model Health Notifications (executing, 5/9 55%)
📋 auth-abc1 — Add Authentication (planning)

STATUS: success

/plan archive <plan-id>

bash "$HOME/.openclaw/scripts/plan-manager.sh" archive "<plan-id>"

Agent Behavior Rules

During Planning Status

• CAN: Read files, query data, search repos, gather context, build plan structure
• CANNOT: Modify files, change configs, create/delete resources, run destructive commands
• MUST: Log all research findings via add-research
• MUST: Update the card after research findings are gathered

During Executing Status

• CAN: Full agent capabilities — read, write, create, modify
• MUST: Mark each step active before starting work
• MUST: Mark each step done after completing work
• MUST: Update the card after each step completion
• MUST: Advance phases when all steps in a phase are done
• SHOULD: Post progress updates in the plan thread

During Gate Status

• CANNOT: Proceed to next phase without Robert's approval
• MUST: Present what was accomplished in the completed phase
• MUST: Wait for Continue/Revise/Stop button click

Cross-Agent Awareness

When a plan involves multiple agents:

• Captain tags related tasks with the plan-id
• Each agent marks their steps done via plan-manager.sh step-done
• Scribe (spec-projects) is the default plan owner
• Other agents are workers — they execute assigned steps

Card Update Pattern

After any state change, always:

1. Run plan-manager.sh card <plan-id> to get updated card JSON
2. Edit the original Discord message with the new embed + components
3. If it's a major state change (approved, phase complete, gate reached), also post a notification in the thread

Storage

Plans stored in ~/.openclaw/plans/<plan-id>.json — managed exclusively by the script. Archived plans move to ~/.openclaw/plans/archive/.

Rules

• Always use the script — never edit plan JSON directly
• One card per plan — edit in place, never recreate
• Thread for all discussion — keep the main channel clean
• Update card on every step completion — Robert should see live progress
• Research findings logged before plan is built — show your work
• If no plan-id specified on sub-commands, use the most recent active plan

Intent: Competent [I03]. Purpose: [P-TBD].