Slash Commands Reference

PD interacts with you through OpenClaw slash commands. This document lists all available commands with plain-language usage instructions.

Reading Guide

New here? Start with the User Guide to understand PD's core concepts, then come back here to look up specific commands.

Quick Reference Table

All commands start with /pd- (short aliases start with /pd + a single letter). They never conflict with OpenClaw's built-in commands.

Command	Short	Category	One-line Purpose
`/pd-init`	`/pdi`	Setup	Initialize workspace, establish project focus
`/pd-bootstrap`	`/pdb`	Setup	Scan local toolchain (rg/sg/fd, etc.)
`/pd-research`	`/pdr`	Setup	Ask Agent to research tool upgrades
`/pd-help`	`/pdh`	Setup	List commands in-session
`/pd-status`	—	Monitoring	View GFI friction index and mental mode
`/pd-pain`	—	Monitoring	Manually report a pain signal
`/pd-evolution-status`	—	Monitoring	View principle evolution loop status
`/pd-context`	—	Config	Control context injection (Thinking OS / project focus)
`/pd-focus`	—	Config	Manage CURRENT_FOCUS.md (compress/rollback)
`/pd-thinking`	`/pdt`	Config	Manage mental models (view/propose/audit)
`/pd-rollback`	—	Rollback	Rollback a misjudged empathy event penalty
`/pd-principle-rollback`	—	Rollback	Rollback a principle and blacklist it
`/pd-export`	—	Data	Export analytics or correction samples
`/pd-samples`	—	Data	View and review correction samples

About Implementation Lifecycle Commands

/pd-promote-impl, /pd-disable-impl, /pd-archive-impl, and /pd-rollback-impl have their replay-generation path retired in PRI-230 and are now in a semi-deprecated state. For the new implementation promotion workflow, use the pd candidate internalize and pd runtime activation promote CLI commands instead. This page no longer documents those four commands.

Setup Commands

`/pd-init`

Short:/pdi

Initializes a new workspace. The command itself only outputs guidance text — the actual initialization is done by the Agent following that guidance: it interviews you to establish your project's strategic focus.

Usage

/pd-init

What Happens

The command tells the Agent to read existing context in the OKR/ directory
The Agent interviews you: What's the project vision? What are the top 1-3 priorities right now?
The Agent generates CURRENT_FOCUS.md (current focus) and USER_CONTEXT.md (user preferences)

When to Use

The first time you use PD in a new workspace
When project direction shifts significantly and you need to realign

TIP

/pd-init does not directly generate PRINCIPLES.md or THINKING_OS.md — those files are created automatically by PD's background services when first needed.

`/pd-bootstrap`

Short:/pdb

Scans your local environment to detect which development tools are installed (rg, sg, fd, qmd, ast-grep, shellcheck) and writes the result to .state/SYSTEM_CAPABILITIES.json.

Usage

/pd-bootstrap

Example Output

🔍 Environment perception complete.
**Detected tools:** `rg`, `fd`, `ast-grep`
**Platform:** darwin
Capabilities saved to `.state/SYSTEM_CAPABILITIES.json`.

When to Use

The first time you use PD after installation
After installing a new tool (e.g., you just installed ast-grep) and want PD to be aware of it

`/pd-research`

Short:/pdr

Asks the Agent to use web search to research the latest tools in a category and output a "Tool Upgrade Proposal". Text-only output, no side effects.

Usage

/pd-research [category]

Parameters

Parameter	Required	Description
`category`	No	Tool category to research; defaults to "modern high-performance CLI tools for coding and architecture"

Examples

/pd-research
/pd-research fast code search tools

When to Use

When you want to learn about the latest tools in a category (code search, doc generation, etc.)
When /pd-bootstrap shows a missing tool and you want alternatives

`/pd-help`

Short:/pdh

Quickly view all PD commands in-session. Since OpenClaw's /help does not automatically list plugin commands, PD provides its own help command.

Usage

/pd-help

TIP

Can't remember the commands? Just type /pdh to see a quick reference table of all commands.

Monitoring Commands

`/pd-status`

View PD's system health. This is one of the commands you'll use most often.

Usage

/pd-status [subcommand]

Subcommands

Subcommand	Description
(none)	Show GFI friction index, mental mode, pain dictionary stats
`empathy [--today\|--week\|--session]`	View empathy event stats (default: today)
`reset`	Reset the current session's GFI to zero
`data`	View trajectory database stats (turns, tool calls, pain events, etc.)

Example Output (default)

📊 Principles Disciple - System Health Monitor
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
💊 Current Friction (GFI): [██████░░░░░░░░░] 35/100
🧠 Current Mental Mode: 🤝 CONCILIATORY
   ↳ Diagnosis: High Friction 🟡

🧠 Evolution Dictionary: 12 active rules
   ↳ Successfully blocked 47 invalid operations
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

What is GFI?

GFI (Global Friction Index) is the current session's "fatigue index". Every time the Agent makes a mistake, gets corrected, or gets blocked, GFI accumulates. The higher the GFI, the more likely the Agent enters "conciliatory mode" — it becomes more cautious, slower, and more apologetic.

0-20: Healthy 🟢
21-50: Minor Issues 🟢
51-80: High Friction 🟡 (consider whether the context is messy)
81-100: Critical 🔴 (consider /pd-status reset or starting a new session)

When to use reset

When the Agent has accumulated too much "emotional baggage" and becomes overly cautious or apologetic, /pd-status reset clears the GFI so the Agent can start fresh. This doesn't "delete memory" — it only resets the friction accumulation.

`/pd-pain`

Manually report a pain signal to PD. PD usually detects pain automatically (tool failures, user corrections, etc.), but sometimes you want to explicitly record an issue — use this command.

Usage

/pd-pain <description of the issue>

Parameters

Parameter	Required	Description
`description`	Yes	Describe the issue in natural language

Example

/pd-pain Agent edited the file without reading it first, overwriting existing logic

Example Output

✅ Pain recorded (context-bound)

📋 Pain ID: manual_1751500000000_a1b2c3d4
📝 Reason: Agent edited the file without reading it first, overwriting existing logic
🔗 Provenance: openclaw_context_bound
📌 Session: sess_xxx

The system will diagnose using current session context.

Difference from Automatic Detection

Automatic detection: PD's hook judges whether something is pain after a tool call (tool failure, blocked, etc.)
/pd-pain: You actively record a "soft" issue (e.g., the Agent's judgment direction was wrong, but no error occurred)

WARNING

/pd-pain must be used inside an OpenClaw chat session — it depends on the current session's context (context-bound provenance). In a non-session environment it will report "Session ID not available".

`/pd-evolution-status`

View the full state of the principle evolution loop: control plane (GFI/gate), evolution plane (queue/tasks), principle stats, workflow funnel.

Usage

/pd-evolution-status

Output Contents

Control Plane: Current GFI, GFI sources, recent gate blocks/bypasses
Evolution: Evolution queue status (pending/in_progress/completed), current evolution task
Principles: Principle stats (candidate/probation/active/archived counts)
Workflow Funnel: Pass rates at each workflow stage

When to Use

When you want to know how many principle candidates are waiting for review
When you want to confirm that evolution tasks are progressing normally
When troubleshooting "why isn't my principle being activated"

Configuration Commands

`/pd-context`

Control what context PD injects into the Agent. This is the main entry point for adjusting PD's "presence".

Usage

/pd-context [subcommand]

Subcommands

Subcommand	Description
`status`	View current injection status (default)
`thinking on\|off`	Toggle Thinking OS injection
`focus full\|summary\|off`	Set project context mode
`minimal`	Preset: core principles only (quietest)
`standard`	Preset: core principles, no Thinking OS
`full`	Preset: core principles + Thinking OS + project context (fullest)
`help`	Show help

Examples

/pd-context status
/pd-context thinking on
/pd-context focus summary
/pd-context full

Preset Differences

Preset	Core Principles	Thinking OS	Project Context	Best For
`minimal`	✅	❌	❌	Minimal interference, let Agent work freely
`standard`	✅	❌	❌	Daily use (same as minimal)
`full`	✅	✅	✅ (summary)	Complex tasks, Agent needs full picture

Core principles are always injected

Regardless of preset, core principles (always-on) are always injected and cannot be disabled. What's configurable is only the "enhancement layers" like Thinking OS and project context.

Where is the config stored?

Configuration is written to .pd/config.yaml (the ADR-0016 unified config file) and takes effect on the next turn.

`/pd-focus`

Manage the CURRENT_FOCUS.md file — this file records the current project focus, current tasks, and next steps. As work progresses, this file grows longer; /pd-focus helps you compress and roll it back.

Usage

/pd-focus [subcommand]

Subcommands

Subcommand	Short	Description
`status`	—	View current status and historical versions
`history`	`hist`	List historical versions
`compress`	`cp`	Manually compress (archive milestones, clean completed items)
`rollback <n>`	`rb <n>`	Rollback to a specified historical version
`help`	—	Show help

Examples

/pd-focus status
/pd-focus compress
/pd-focus rollback 3

When to Use compress

When CURRENT_FOCUS.md exceeds 40 lines
When you've completed a milestone and want to clear completed items while keeping unfinished ones
When you want to archive milestones to the memory/ diary

Auto-compression

PD also auto-compresses CURRENT_FOCUS.md in the background. /pd-focus compress is a manual trigger for when you want to clean up immediately.

`/pd-thinking`

Short:/pdt

Manage mental models (Thinking OS). Mental models are PD's "cognitive layer" — they tell the Agent which thinking style to use in different scenarios.

Off by default

Thinking OS is off by default. To enable it, first run /pd-context thinking on, then use /pd-thinking to view stats.

Usage

/pd-thinking [subcommand]

Subcommands

Subcommand	Description
`status`	View usage stats for each mental model (hit rate, dormant models)
`propose <description>`	Propose a new mental model (must include a "signal detection" section)
`audit`	Audit model health (over-triggered/dormant/pending candidates)
(none)	Show help

Examples

/pd-thinking status
/pd-thinking propose A mental model for code review, signal detection: regex /review|audit/
/pd-thinking audit

propose Requirements

A proposal must include the keyword "signal" and describe how to detect the model's usage via regex. Proposals are written to THINKING_OS_CANDIDATES.md and require validation in 3+ different task types and human approval before promotion.

What audit Tells You

✅ Healthy models (reasonable hit rate)
🔸 Over-triggered models (hit rate >50%, pattern may be too broad)
⚠️ Dormant models (0 hits, consider archiving)
📝 Number of pending candidates

Rollback Commands

PD's core promise is "reversibility". These commands let you undo PD's decisions.

`/pd-rollback`

Rollback the penalty from an empathy event. When PD misjudges a "user frustration" and GFI rises incorrectly, use this command to undo it.

Usage

/pd-rollback <event-id>
/pd-rollback last

Parameters

Parameter	Required	Description
`event-id`	Yes	The event ID to rollback (or use `last` for the most recent)

Examples

/pd-rollback last
/pd-rollback evt_20260701_001

last vs <event-id>

last: Rollback the most recent empathy event in the current session (most common)
<event-id>: Rollback a specific event (get the ID from /pd-status empathy output)

What Happens After Rollback

The event's penalty score is subtracted from GFI
The event is marked as "rolled back"
The Agent's friction level decreases accordingly

WARNING

/pd-rollback must be used in a chat session (depends on sessionId). It rolls back "empathy penalty", not "principles" — to rollback a principle, use /pd-principle-rollback.

`/pd-principle-rollback`

Rollback a principle and add its pattern to the blacklist (preventing it from being re-proposed).

Usage

/pd-principle-rollback <principle-id> [reason]

Parameters

Parameter	Required	Description
`principle-id`	Yes	Principle ID (e.g., `P_001`, from `/pd-evolution-status` or `PRINCIPLES.md`)
`reason`	No	Reason for rollback (default: "manual rollback")

Examples

/pd-principle-rollback P_003
/pd-principle-rollback P_003 This principle is too strict and blocks normal development

Difference from /pd-rollback

Command	Rollback Target	Effect
`/pd-rollback`	One empathy event	GFI decreases, principles unaffected
`/pd-principle-rollback`	One principle	Principle deactivated + pattern blacklisted, prevents re-proposal

Data Commands

`/pd-export`

Export PD's collected data for analysis or backup.

Usage

/pd-export analytics
/pd-export corrections [--redacted]

Subcommands

Subcommand	Description
`analytics`	Export analytics snapshot (aggregated stats)
`corrections [--redacted]`	Export correction samples (approved ones)

What is --redacted?

The --redacted flag applies redaction (removes sensitive info), suitable for sharing or reporting. Without it, raw data is exported.

Examples

/pd-export analytics
/pd-export corrections --redacted

Example Output

Exported correction samples to .state/exports/corrections_20260701.json (mode=redacted, count=15).

`/pd-samples`

View and review correction samples. Correction samples are "training data" extracted by PD from the Agent's mistakes.

Usage

/pd-samples
/pd-samples review approve <sample-id> [note]
/pd-samples review reject <sample-id> [note]

Subcommands

Subcommand	Description
(none)	List all pending samples
`review approve <id> [note]`	Approve a sample
`review reject <id> [note]`	Reject a sample

Examples

/pd-samples
/pd-samples review approve s_001 good quality
/pd-samples review reject s_002 this isn't actually an error

What reject Triggers

Rejecting a sample isn't just "not approved" — it triggers a correction_rejected pain event, causing PD to re-diagnose. This is PD's "reverse learning" mechanism: when you reject a wrong correction sample, PD learns that its judgment was off.

Review Workflow

/pd-samples to see the pending list
Decide approve or reject based on sample quality
Approved samples enter training data; rejected samples trigger re-diagnosis

Common Workflows

Workflow 1: First-time PD Setup

1. /pdi              # Initialize workspace, establish project focus
2. /pdb              # Scan local tools
3. /pd-status        # Confirm PD is running normally
4. /pd-context full  # Enable full context (optional, for complex projects)

Workflow 2: Correcting Agent Mistakes

1. /pd-pain Agent edited without reading the file first, overwrote my code
2. /pd-status        # Check if GFI rose
3. (Wait for PD background diagnosis, or use /pd-evolution-status to view queue)

Workflow 3: Agent is Over-fatigued

When the Agent keeps apologizing and is overly cautious:

1. /pd-status        # Check GFI, confirm it's too high
2. /pd-status reset  # Clear GFI
3. (Continue working, Agent returns to efficient mode)

Workflow 4: Rollback a Misjudgment

PD misjudged a "user frustration" and GFI rose incorrectly:

1. /pd-status empathy --session  # View current session's empathy events
2. /pd-rollback last             # Rollback the most recent misjudgment
3. /pd-status                    # Confirm GFI has decreased

Workflow 5: Review Correction Samples

1. /pd-samples                    # View pending samples
2. /pd-samples review approve s_001  # Approve good samples
3. /pd-samples review reject s_002   # Reject wrong samples (triggers re-diagnosis)
4. /pd-export corrections --redacted  # Export redacted data (optional)

Workflow 6: Principle Governance

1. /pd-evolution-status           # See what candidate principles exist
2. (Review principle candidates in the console)
3. /pd-principle-rollback P_003   # Rollback an inappropriate activated principle

FAQ

Command not responding?

Confirm PD plugin is installed: if /pd-help responds, the plugin loaded successfully
Confirm you're in a workspace: PD commands need workspace context
Check logs: ~/.openclaw/workspace/memory/logs/SYSTEM_*.log

`/pd-pain` says "Session ID not available"?

/pd-pain must be used inside an OpenClaw chat session, not from the CLI or a non-session environment.

Output is in English, how do I switch to Chinese?

Edit your workspace's .pd/config.yaml and set language: zh. Or set it in the plugin config.

GFI won't go down?

Use /pd-status reset to clear the current session's GFI
Check if there are persistent tool failures triggering pain (fix the root cause)
Consider /clear to start a new session

How do I view all activated principles?

Run /pd-evolution-status for stats
Read ~/.openclaw/workspace/.principles/PRINCIPLES.md directly

User Guide — PD core concepts and workflows
Getting Started — Installation and first-time setup
Development Guide — The pd CLI tool for out-of-session use

Slash Commands Reference ​

Quick Reference Table ​

Setup Commands ​

/pd-init ​

/pd-bootstrap ​

/pd-research ​

/pd-help ​

Monitoring Commands ​

/pd-status ​

/pd-pain ​

/pd-evolution-status ​

Configuration Commands ​

/pd-context ​

/pd-focus ​

/pd-thinking ​

Rollback Commands ​

/pd-rollback ​

/pd-principle-rollback ​

Data Commands ​

/pd-export ​

/pd-samples ​

Common Workflows ​

Workflow 1: First-time PD Setup ​

Workflow 2: Correcting Agent Mistakes ​

Workflow 3: Agent is Over-fatigued ​

Workflow 4: Rollback a Misjudgment ​

Workflow 5: Review Correction Samples ​

Workflow 6: Principle Governance ​

FAQ ​

Command not responding? ​

/pd-pain says "Session ID not available"? ​

Output is in English, how do I switch to Chinese? ​

GFI won't go down? ​

How do I view all activated principles? ​

Related Documentation ​

Slash Commands Reference

Quick Reference Table

Setup Commands

`/pd-init`

`/pd-bootstrap`

`/pd-research`

`/pd-help`

Monitoring Commands

`/pd-status`

`/pd-pain`

`/pd-evolution-status`

Configuration Commands

`/pd-context`

`/pd-focus`

`/pd-thinking`

Rollback Commands

`/pd-rollback`

`/pd-principle-rollback`

Data Commands

`/pd-export`

`/pd-samples`

Common Workflows

Workflow 1: First-time PD Setup

Workflow 2: Correcting Agent Mistakes

Workflow 3: Agent is Over-fatigued

Workflow 4: Rollback a Misjudgment

Workflow 5: Review Correction Samples

Workflow 6: Principle Governance

FAQ

Command not responding?

`/pd-pain` says "Session ID not available"?

Output is in English, how do I switch to Chinese?

GFI won't go down?

How do I view all activated principles?

Related Documentation