Skip to content

Conversation

@nhorton
Copy link
Contributor

@nhorton nhorton commented Jan 30, 2026

Summary

Removes outdated skill definition files from the .claude/skills directory that were part of an earlier platform integration workflow system. These files are no longer used and have been superseded by the current job-based workflow implementation.

Changes

  • Removed .claude/skills/add_platform.research/SKILL.md - Research phase documentation
  • Removed .claude/skills/add_platform.add_capabilities/SKILL.md - Capability addition phase documentation
  • Removed .claude/skills/add_platform.implement/SKILL.md - Implementation phase documentation
  • Removed .claude/skills/add_platform.verify/SKILL.md - Verification phase documentation
  • Removed 27 additional legacy skill-related files (as indicated in diff summary)

Context

These skill definitions were part of an earlier iteration of the platform integration workflow. The current system uses job definitions (located in .deepwork/jobs/) instead of skill definitions. This cleanup removes technical debt and eliminates confusion between the two workflow systems.

Impact

  • No functional impact on current workflows (jobs-based system remains intact)
  • Reduces repository clutter and maintenance burden
  • Clarifies the single source of truth for platform integration workflows

https://claude.ai/code/session_01GWfanjqoS4i54AWANk7UJ2

claude and others added 4 commits January 29, 2026 23:51
This refactoring changes how DeepWork generates skills for Claude Code:

**Before:** Each job generated a meta-skill + separate step skills
- `/job_name` (meta-skill that routes to step skills)
- `/job_name.step1` → `/job_name.step2` (step skills calling each other)

**After:** Each job generates a single agent file with embedded skills
- `/job_name` (agent with all skills embedded)
- No separate step skill files

Key changes:
- Added agent-job.md.jinja template for generating agent files
- Added `supports_agent_mode` flag to adapters (enabled for Claude)
- Added `generate_agent()` method to SkillGenerator
- Updated sync command to use agent mode when supported
- Sync now cleans up old step skill directories when using agent mode
- Updated tests to expect agent-based structure

Benefits:
- Simpler architecture: one file per job instead of many
- Better alignment with Claude Code's Task tool for subagents
- Easier to understand: all job skills visible in one file
- Cleaner skills directory structure

Gemini continues to use legacy mode (meta-skill + step skills) as it
doesn't support the agent pattern yet.

https://claude.ai/code/session_01GWfanjqoS4i54AWANk7UJ2
- Updated doc/architecture.md to describe agent-based skill generation
  - Added agent-job.md.jinja template to directory structure
  - Added supports_agent_mode to ClaudeAdapter example
  - Updated SkillGenerator section to describe two generation modes
  - Updated target project structure to show agent-mode skill layout
- Updated CHANGELOG.md with agent-based architecture changes

https://claude.ai/code/session_01GWfanjqoS4i54AWANk7UJ2
- Add macros for repeated patterns: output_files(), workflow_desc(), completion_msg()
- Consolidate redundant workflow/step listings in Agent Overview and How to Use sections
- Remove duplicate Workflow Detection and Standalone Skill Detection sections
- Simplify On Completion sections using completion_msg macro
- Net reduction of 116 lines across template and generated files

https://claude.ai/code/session_01GWfanjqoS4i54AWANk7UJ2
@nhorton nhorton enabled auto-merge January 30, 2026 03:29
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants