feat: Add extensible custom slash command support #50

robotdad · 2026-01-13T23:17:47Z

Summary

Adds support for extensible custom slash commands in the interactive CLI, enabling users to define reusable prompt templates as /commands.

Changes

Core Integration (main.py)

CommandProcessor now loads custom commands from slash_command tool at startup
Custom commands appear in /help under "Custom Commands:" section
/reload-commands built-in command to refresh commands without restart
Template execution returns substituted prompt to REPL for LLM processing

Behavior Composition (config.py)

New _build_cli_behaviors() consolidates CLI-specific behaviors
Always composes slash-command behavior for interactive sessions
Refactored notification behaviors to be called from new function

Documentation (README.md)

Quick start guide for creating custom commands
Command locations and precedence rules
Template syntax documentation

How It Works

Users create .md files in .amplifier/commands/ or ~/.amplifier/commands/
Files define prompt templates with $ARGUMENTS, $1, $2, etc.
Typing /commandname args substitutes and executes the template

Example

<!-- ~/.amplifier/commands/review.md -->
---
name: review
description: Code review for a file
---
Review this file for bugs, security issues, and improvements: $ARGUMENTS

Usage: /review src/auth.py

Dependencies

Requires amplifier-module-tool-slash-command
PR pending: robotdad/amplifier-module-tool-slash-command#feat/add-behavior-bundle

Testing

Shadow environment validation confirmed feature branch structure
Existing unit tests pass

🤖 Generated with Amplifier

Co-Authored-By: Amplifier 240397093+microsoft-amplifier@users.noreply.github.com

Integrate with slash_command module to support user-defined commands: - CommandProcessor now loads custom commands from slash_command tool - Custom commands appear in /help alongside built-in commands - Add /reload-commands to refresh commands from disk - Custom command templates are substituted and executed as prompts Commands are discovered from: - .amplifier/commands/ (project-level) - ~/.amplifier/commands/ (user-level) Requires: robotdad/amplifier-module-tool-slash-command

Fixes 'No module named pygments.lexers.markup' error. The markup lexers module requires Pygments 2.12+.

Documents: - Quick start for creating custom commands - Command locations and precedence - Template syntax (, , {{default}}) - Built-in commands (/help, /reload-commands) - Link to full documentation

Adds slash-command behavior to CLI's compose_behaviors, enabling extensible /commands for interactive sessions. Changes: - Add _build_cli_behaviors() to consolidate CLI-specific behaviors - Always compose slash-command behavior for interactive CLI - Refactor to call _build_notification_behaviors() from new function The slash_command tool is now automatically available, providing: - Custom commands from .amplifier/commands/ (project) and ~/.amplifier/commands/ (user) - Template substitution with $ARGUMENTS, $1, $2, etc. - Integration with existing CommandProcessor in main.py 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

robotdad · 2026-01-22T19:25:17Z

✅ Smoke Test Validation

This change has been validated with the amplifier-bundle-smoke-test recipe.

Results

Phase	Status
CLI Core	✅ PASS
CLI Config	✅ PASS
CLI Resources	✅ PASS
Recipe Validation	✅ PASS
Source Override	✅ PASS
Multi-Provider	✅ PASS (anthropic, openai)
Session CRUD	✅ PASS
Mention Validation	✅ PASS
Provider Responds	✅ PASS
Tool Execution	✅ PASS
Agent Delegation	✅ PASS

Summary: 11/11 core tests passed

Note: Ollama and Azure OpenAI were skipped (not configured in test environment) - this is expected behavior.

🤖 Validated with Amplifier smoke-test bundle

The CommandExecutor.execute() method is async but was being called without await, causing the coroutine object to be returned instead of the actual result. 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

registry.get_command_dict() already returns keys with / prefix, so adding another / resulted in //review instead of /review. 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

robotdad · 2026-01-22T23:53:39Z

Smoke Test Results (Post-Changes)

All core functionality passed. Summary:

Test	Result
CLI Core	✅ PASS
CLI Config	✅ PASS
CLI Resources	✅ PASS
Recipe Validation	✅ PASS
Source Override	✅ PASS
Session CRUD	✅ PASS
Mention Validation	✅ PASS
Provider Responds	✅ PASS
Tool Execution	✅ PASS
Agent Delegation	✅ PASS

Multi-Provider

Provider	Result
OpenAI	✅ PASS
Anthropic	✅ PASS
Azure OpenAI	⏭️ SKIPPED (not configured)
Ollama	⏭️ SKIPPED (not configured)

Summary: 12 passed, 0 failed, 2 skipped

Verdict: ✅ PASS - All configured providers working correctly. Azure and Ollama skipped as expected (not configured on test machine).

Add support for the custom slash commands directory (~/.amplifier/commands/) in the reset command: - Add "commands" to RESET_CATEGORIES mapping - Include in CATEGORY_ORDER between keys and cache - Add description for interactive UI display - Add to DEFAULT_PRESERVE set (user content preserved by default) - Update docstring and help text examples This ensures user-created custom commands are recognized by reset and protected from accidental deletion unless explicitly requested. 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

Refactor to follow upstream's pattern of separate behavior builder functions for each category, making future syncs easier. Before: A unified _build_cli_behaviors() function returned all CLI behaviors (modes, slash-commands, notifications) in one list. After: Separate functions following upstream's convention: - _build_modes_behaviors() - returns modes behavior URI - _build_slash_command_behaviors() - returns slash command behavior URI - _build_notification_behaviors() - already existed, unchanged The caller in resolve_bundle_config() now uses extend() calls for each function, matching upstream's architectural pattern. 🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier) Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>

robotdad · 2026-01-23T02:59:17Z

Today's Changes Summary

The following commits were added today to address issues discovered during testing and improve code organization:

Bug Fixes

fix: await async executor.execute (6e94f88) - Fixed missing await on the async CommandExecutor.execute() method, which was causing coroutine objects to be returned instead of actual results
fix: remove duplicate slash prefix (71928c7) - Fixed double-slash issue where commands appeared as //review instead of /review since registry.get_command_dict() already returns keys with the / prefix

Features

feat: add commands category to reset command (43b3c85) - Added recognition of the custom slash commands directory (~/.amplifier/commands/) in the reset command, with appropriate protection from accidental deletion

Refactoring

refactor: split _build_cli_behaviors (5f97475) - Split unified _build_cli_behaviors() into separate builder functions (_build_modes_behaviors(), _build_slash_command_behaviors()) to follow upstream's architectural pattern and ease future syncs

Smoke Test Results

============================================================
AMPLIFIER SMOKE TEST RESULTS
============================================================

CLI Core:                 ✅ PASS
CLI Config:               ✅ PASS
CLI Resources:            ✅ PASS
Recipe Validation:        ✅ PASS
Source Override:          ✅ PASS
Multi-Provider:           ⏭️ SKIP (Ollama not configured)
Session CRUD:             ✅ PASS
Mention Validation:       ✅ PASS
Provider Responds:        ✅ PASS
Tool Execution:           ✅ PASS
Agent Delegation:         ✅ PASS

------------------------------------------------------------
SUMMARY: 10 passed, 0 failed, 1 skipped (of 11)
------------------------------------------------------------

SMOKE TEST: PASS ✅

Note: The Multi-Provider skip is due to Ollama not being configured on the test machine. Both OpenAI and Anthropic providers passed successfully.

robotdad mentioned this pull request Jan 13, 2026

feat: add behavior bundle for CLI integration microsoft/amplifier-module-tool-slash-command#5

Merged

robotdad and others added 4 commits January 22, 2026 10:18

fix: Add explicit pygments>=2.15.0 dependency

aff82fc

Fixes 'No module named pygments.lexers.markup' error. The markup lexers module requires Pygments 2.12+.

docs: Add custom slash commands section to README

353f6b9

Documents: - Quick start for creating custom commands - Command locations and precedence - Template syntax (, , {{default}}) - Built-in commands (/help, /reload-commands) - Link to full documentation

robotdad force-pushed the feat/custom-slash-commands branch from 3a98864 to 879a005 Compare January 22, 2026 18:19

robotdad and others added 2 commits January 22, 2026 13:36

robotdad and others added 2 commits January 22, 2026 16:12

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

feat: Add extensible custom slash command support #50

feat: Add extensible custom slash command support #50

robotdad commented Jan 13, 2026

Uh oh!

robotdad commented Jan 22, 2026

Uh oh!

robotdad commented Jan 22, 2026

Uh oh!

robotdad commented Jan 23, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

feat: Add extensible custom slash command support #50

Are you sure you want to change the base?

feat: Add extensible custom slash command support #50

Conversation

robotdad commented Jan 13, 2026

Summary

Changes

Core Integration (main.py)

Behavior Composition (config.py)

Documentation (README.md)

How It Works

Example

Dependencies

Testing

Uh oh!

robotdad commented Jan 22, 2026

✅ Smoke Test Validation

Results

Uh oh!

robotdad commented Jan 22, 2026

Smoke Test Results (Post-Changes)

Multi-Provider

Uh oh!

robotdad commented Jan 23, 2026

Today's Changes Summary

Bug Fixes

Features

Refactoring

Smoke Test Results

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant