skills.rest
alsdmlals4-eng
alsdmlals4-engCOMMUNITY
Explore all skills by this author →
GitHub

orchestrating-deepseek-worktrees

Isolates external AI drafting tasks in dedicated git worktrees and reviews outputs before merging.

Updated Jul 9, 2026
Search Tags:#code review#orchestration#git worktree#handoff#deepseek#branch isolation#external ai
One-click install
npx skills add https://github.com/alsdmlals4-eng/Base --skill orchestrating-deepseek-worktrees-alsdmlals4-eng
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: orchestrating-deepseek-worktrees
Source: https://github.com/alsdmlals4-eng/Base/tree/main/skills/orchestrating-deepseek-worktrees
Command: npx skills add https://github.com/alsdmlals4-eng/Base --skill orchestrating-deepseek-worktrees-alsdmlals4-eng

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Delegating large drafting, classification, or comparison tasks to an external AI model risks contaminating the main repository with unreviewed output. This Skill provides a governed workflow that isolates external AI work in separate git worktrees and branches, treats all results as REVIEW_PENDING, and enforces rehydration of current repository canon before any integration. ## Core Features & Use Cases - Isolated Worktree Orchestration: Creates dedicated .worktrees/deepseek-<topic>/ directories and ai/deepseek-<topic> branches so external AI drafts never touch the main worktree directly. - Executor Rehydration Gate: Requires any external executor to re-read current AGENTS.md, exact commit SHA, allowlists, and protected paths before acting, preventing stale handoff packages from becoming false authority. - Structured Review Contract: Returns drafts with separated evidence, assumptions, and unverified items so the current authorized Work can validate diffs against repository canon before merging. - Use Case: You need to classify hundreds of data cards or draft a long planning document. Package the task with an allowlisted context, let an external model produce candidates in an isolated branch, then review and merge only the approved portions. ## Quick Start Ask the AI to isolate a large document drafting task into a DeepSeek worktree, generate a review-pending draft, and validate it against the current repository canon before merging.

Frequently Asked Questions about orchestrating-deepseek-worktrees

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I safely use an external AI model for large drafting tasks in a git repository?▼

Isolate the task in a dedicated worktree such as .worktrees/deepseek-<topic>/ with its own ai/deepseek-<topic> branch, provide only allowlisted context, and treat all output as REVIEW_PENDING. The current authorized Work reviews diffs against repository canon before merging anything.

What is an executor rehydration gate in AI handoff workflows?▼

It is a mandatory checkpoint where the receiving executor re-reads the latest user instructions, AGENTS.md, exact commit SHA, allowlists, protected paths, and current worktree state instead of trusting a handoff summary. Current truth always overrides stale handoff content.

When should I not delegate work to an external AI model?▼

Avoid delegation for final decisions on security, payments, or destructive migrations, for final approval of real product bugs, when secrets or unauthorized private data would be transmitted, or when a task is small and delegation is only a workaround.

Can external AI results be merged directly into the main branch?▼

No. External AI output stays in an isolated branch as REVIEW_PENDING input. The current authorized Work must validate the diff, evidence, and assumptions against repository canon, then implement only the approved minimal changes.

What happens if runtime validation is unavailable during delegated work?▼

The missing runtime verification is recorded as NOT_RUN with an explicit evidence ceiling. Independent approved implementation and static checks continue, but full completion or release PASS cannot be claimed without the required runtime evidence.

Related Skills

Explore More Software Engineering Skills→
tanveerriaztanveerriaz
1

git-release-hygiene

Enforces working-branch-only commits, conventional messages, and approval-gated merges to protected main branches.

Community
Intermediate
zcawood-miezcawood-mie

code-cleanup

Clean up committed code changes by minimizing diffs and running linters.

Community
Intermediate
synthaicodesynthaicode
4

judgment_log

Writes structured judgment logs recording decisions, evidence, confidence, and next verification steps.

Community
Intermediate
chadyi-kingchadyi-king

task-decomposer

Decompose complex requests into executable subtasks and map capabilities.

Community
Advanced