Rename commands to skills throughout CI/CD workflows by nhorton · Pull Request #179 · Unsupervisedcom/deepwork

nhorton · 2026-01-29T16:23:14Z

Summary

This PR updates all references from "commands" to "skills" across the CI/CD workflows and documentation to align with the project's terminology update.

Key Changes

Updated workflow documentation in .github/workflows/README.md to reflect the new "skill generation" terminology
Renamed all file paths from .claude/commands/ to .claude/skills/ in the integration test workflow
Updated test validation logic to check for SKILL.md files in skill directories instead of .md files in a flat commands directory
Updated all echo messages, comments, and step names to use "skill" instead of "command"
Updated artifact upload paths to reference the new skills directory structure

Notable Implementation Details

The skill directory structure now uses subdirectories (e.g., fruits.identify/SKILL.md) instead of flat files (e.g., fruits.identify.md)
Meta-skills for jobs are now stored in their own directories (e.g., fruits/SKILL.md)
All validation checks in the test workflow have been updated to verify the new directory structure
Both the fixture-based validation job and the full end-to-end test job have been updated consistently

https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

The validate-generation CI job was looking for `.claude/commands/` which is the old structure. DeepWork now generates skills in `.claude/skills/` with SKILL.md files inside directories. Changes: - Update validate-generation job to check for skill directories - Update claude-code-e2e job to use skill paths - Update artifact upload paths for new skill structure - Update README documentation to reference skills https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

…ocation - Rename TestCommandGenerationE2E to TestSkillGenerationE2E - Rename project_with_commands fixture to project_with_skills - Fix claude invocation: use /fruits instead of appending args to skill name - Provide input via stdin instead of as part of skill path - Update docstrings from "commands" to "skills" https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

The --yes flag is not a valid claude CLI option. Remove it from all subprocess.run calls that invoke claude. https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

- Remove invalid --yes flag from all claude CLI invocations - Consolidate STEP 3 and STEP 4 into single /fruits workflow invocation - The meta-skill automatically runs all sub-steps (identify + classify) - Update timeout to 10 minutes for full workflow execution - Update summary to reflect the simplified workflow https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

Replace three separate tests (test_identify_step_execution, test_classify_step_execution, test_full_workflow_execution) with a single test_fruits_workflow_execution that invokes /fruits once. The /fruits meta-skill automatically runs all sub-steps, so individual step invocations are not needed. This matches how DeepWork is actually used. https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

The e2e test was failing because it relied on Claude to generate a valid job.yml, which is flaky (Claude sometimes adds invalid fields like 'parameters'). Simplify the e2e test to: 1. Copy the pre-existing fruits job fixture into the test project 2. Run deepwork sync to generate skills 3. Execute /fruits workflow and validate outputs This makes the test deterministic while still validating that generated skills work correctly when executed with Claude Code. https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

…ests" This reverts commit e277166.

The e2e test was failing because Claude generated a job.yml with an invalid 'parameters' field at the root level. The job schema only allows: name, version, summary, description, changelog, steps. Update the prompt to explicitly tell Claude: - What root-level fields are allowed - NOT to add fields like 'parameters' or 'config' - The correct format for step inputs https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

Create a .claude/settings.json with allow permissions for: - Bash(*) - Read(./**) - Edit(./**) - Write(./**) - Skill(*) This allows Claude to create output files without permission prompts during CI e2e tests, while still preserving hook functionality. https://claude.ai/code/session_01UUxYSXJyUU8uinbBwNwG84

nhorton enabled auto-merge January 29, 2026 16:24