CAPABILITY/SKILLS/agents/workspace-isolation/SKILL.md
Enforces disciplined git workflow for agents to prevent commit bloat and ensure human approval before permanent repository changes.
npx skillsauth add reneromero08/agent-governance-system workspace-isolationInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
required_canon_version: >=3.0.0
Version: 1.0.0
Status: Active
Enforces disciplined git workflow for agents to prevent commit bloat and ensure human approval before permanent repository changes.
This skill applies to ALL agent sessions that involve repository modifications. Agents MUST follow these rules whenever performing git operations.
The user MUST review and approve ALL work BEFORE any commit or merge.
Agents do NOT decide when work is ready. The human does. Present your work, then STOP and WAIT.
git add to stage, but DO NOT commitgit diff --staged outputCRITICAL: NEVER use git merge or git merge --no-ff. Always use selective file checkout.
# Switch to target branch (e.g., main)
git checkout main
# Cherry-pick ONLY your changed files from feature branch
git checkout feature-branch -- file1.py file2.py file3.md
# Commit the changes
git commit -m "feat: descriptive message"
git merge feature-branchgit merge --no-ff feature-branch# From main repo root (not worktree)
python CAPABILITY/SKILLS/agents/workspace-isolation/run.py cleanup <task_id>
git worktree remove <path> && git branch -D feature-branchAmend is SAFE only when:
Amend is UNSAFE when:
For post-merge cleanup (changelog updates, version bumps):
git commit --amend to add to previous commitAgents MUST NOT:
git commit before all work is completegit commit without user approvalgit merge or git merge --no-ff (use selective file checkout instead)git push without user approvalgit commit --amend on pushed commits without explicit user consent| Action | Requires Approval |
|--------|-------------------|
| git add | No |
| git commit | YES |
| git merge | YES |
| git push | YES |
| git commit --amend | YES (with safety check) |
| git rebase | YES |
| git branch -d | No (auto after merge) |
1. Agent receives task: "Add feature X and update changelog"
2. Agent implements feature X (no commits)
3. Agent updates changelog (no commits)
4. Agent stages all changes: git add -A
5. Agent: "All work complete. Here's what I've done:
[git diff --staged output]
Summary:
- Added feature X in src/feature.ts (new endpoint for user preferences)
- Updated CHANGELOG.md (added entry for v1.2.0)
Proposed commit message: 'feat: add user preferences endpoint'
Please review the changes above. Let me know when you'd like me to commit."
>>> AGENT STOPS HERE AND WAITS <<<
6. User reviews the diff, maybe asks questions, requests changes
7. Agent makes any requested changes, re-stages, presents again
8. User: "Commit it" (explicit approval)
9. Agent: git commit -m "feat: add user preferences endpoint"
10. Agent: "Committed. Ready to merge to main?"
>>> AGENT STOPS HERE AND WAITS <<<
11. User: "Yes, merge"
12. Agent:
```bash
git checkout main
git checkout feature-branch -- src/feature.ts CHANGELOG.md
git commit -m "feat: add user preferences endpoint"
python CAPABILITY/SKILLS/agents/workspace-isolation/run.py cleanup feature-branch
```
13. Agent: "Merged only changed files and cleaned up feature-branch."
14. If post-merge cleanup needed:
Agent: "Need to update version number. Safe to amend? (not pushed yet)"
>>> AGENT STOPS HERE AND WAITS <<<
git diff --staged or git status before requesting commit approvalAgent should output at each checkpoint:
required_canon_version: >=3.0.0
development
<!-- CONTENT_HASH: b22de257a144f83b390075074e3d6a4552ecbc1ec53fc5d4960c6e76dc9807dd --> # Skill: swarm-orchestrator **Version:** 0.1.0 **Status:** Active **required_canon_version:** ">=3.0.0 <4.0.0" **canon_version:** "3.0.0" # Swarm Orchestrator Launches and coordinates Governor + Ant Workers. ## Usage ```bash # Launch Governor python scripts/poll_and_execute.py --role Governor # Launch Ant Workers python scripts/poll_and_execute.py --role Ant-1 python scripts/poll_and_execute.py --role A
tools
<!-- CONTENT_HASH: 9bd07a4ed03d6eea49673a43682a62ca7df8b0cabb7ff424ac91cd35e9e7bea7 --> # Skill: swarm-directive **Version:** 0.1.0 **Status:** Draft **required_canon_version:** ">=2.8.0 <3.0.0" # Swarm Directive Skill Send tasks to your CATALYTIC-DPT swarm from Claude Code, Kilo CLI, or Cline CLI. ## Quick Start ### Option 1: Direct CLI Command (Simplest) ```bash cd "d:\CCC 2.0\AI\agent-governance-system" # Create input cat > /tmp/swarm_task.json << 'EOF' { "directive": "Analyze the
tools
<!-- CONTENT_HASH: 0f69dfe537f3e6cd8b1fdb353c9ab7c031ef8b08379fde8a98c137bb604d799d --> # Qwen CLI - Local AI Assistant **Version:** 1.0.0 **Status:** Active **Required_Canon_Version:** >=2.0.0 **Purpose**: Provides a local CLI interface to Qwen 7B via Ollama for fast, offline AI assistance. **Model**: Qwen2.5 7B (via Ollama) **Use Cases**: - Quick code questions without cloud API costs - Offline development assistance - Fast prototyping and testing - Private/sensitive code analysis ## F
development
<!-- CONTENT_HASH: 7b643dca3adc4d38de7030cb1e962a5130ea4b36ae04ad4bf97ed8889de2a3dc --> # Skill: governor **Version:** 0.1.0 **Status:** Active **required_canon_version:** ">=3.0.0 <4.0.0" **canon_version:** "3.0.0" # Governor The Conductor - analyzes, decomposes, and dispatches tasks to Ant Workers. ## Usage ```bash python scripts/run.py input.json output.json ``` ## Input Schema ```json { "gemini_prompt": "Analyze D:/path/to/files and summarize", "task_id": "analyze-001", "command