# Orchestrate Tasks Across AI Coding Providers

Coordinating several AI coding providers creates routing, isolation, and result collection work. This skill supplies documented workflows for invoking Codex, Gemini, Cursor, and Claude task agents.

## Install

```bash
npx skillstore add consiliency/orchestration
```

## Metadata

- Status: approved
- Slug: consiliency-orchestration
- Skillstore revision: r1
- Version status: missing
- Tree hash: 945ca73ec7b96181006c2b9947c8a15f6dbc5b39162c87b05fdda8378cb19733
- Author: Consiliency
- GitHub username: Consiliency
- License: MIT
- Repository: https://github.com/Consiliency/treesitter-chunker/tree/main/.ai-dev-kit/skills/orchestration
- Ref: 635f69fb8d2f4e6330ba47a4e5a0fb239c04d110
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, filesystem
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/consiliency-orchestration
- Manifest: https://skillstore.pages.dev/api/skills/consiliency-orchestration/manifest

## Capabilities

- Routes work among Codex, Gemini, Cursor, and Claude using task characteristics and provider strengths.
- Documents read-only and write-mode CLI invocation patterns for supported external providers.
- Launches provider tasks in the background and collects results through TaskOutput.
- Defines fallback handling for authentication failures using interactive terminal sessions.
- Recommends isolated Git worktrees for provider tasks requiring file changes.
- Provides routing priorities and fallback chains for analysis, code execution, research, and quick generation.

## Use Cases

- Run Parallel Code Reviews: Send the same repository question to several providers, collect independent findings, and compare their conclusions.
- Design Provider Fallbacks: Choose providers by context size, sandbox needs, speed, and task complexity, then define recovery paths for failures.
- Isolate Delegated Changes: Place write-capable provider work in separate Git worktrees, inspect diffs, and merge only approved changes.

## Prompt Templates

### Route a Read-Only Review

```
Review [repository or files] using [provider]. Keep the provider read-only, run its help command first, and return findings without changing files.
```

### Compare Two Providers

```
Run independent read-only reviews with Codex and Gemini for [task]. Collect both results, identify agreements and conflicts, then provide a concise synthesis.
```

### Isolate a Write Task

```
Create an isolated worktree for [task]. Delegate implementation to [provider], show the diff, run [tests], and wait for approval before merging or cleanup.
```

### Build a Resilient Provider Chain

```
Assess [task] by complexity, context size, sandbox needs, and budget. Select a primary provider, define fallbacks, run suitable tasks in parallel, and synthesize results.
```

## Limitations

- This is documentation only and does not include executable orchestration scripts.
- It requires installed provider CLIs, authenticated accounts, and Claude Code permission configuration.
- Hard-coded provider models and flags may become outdated and must be verified with each CLI help output.
- Write modes and automatic approval flags can modify files, so delegated work requires review and isolation.

## Best Practices

- Verify every provider CLI and current flags with its help output before invocation.
- Keep read-only mode as default and require explicit approval before any write-capable or automatic-approval mode.
- Use unique isolated worktrees, review diffs, and record exact cleanup targets before merging changes.

## Anti Patterns

- Do not grant broad persistent CLI permissions when narrower, task-specific approval is available.
- Do not place untrusted text directly into shell command strings or provider prompts without clear boundaries.
- Do not delete files by wildcard or merge provider branches without reviewing ownership and changes.

## Security Audit

- Audited at: 2026-07-19T03:56:34.071\+00:00
- Summary: Most static alerts are Markdown backticks or fixed documentation paths, but several examples authorize autonomous external commands and write-capable agents. Semantic review also found prompt interpolation, external data disclosure, unscanned instruction dependencies, and unsafe wildcard cleanup.

## Stats

- Views: 378
- Downloads: 5
- Favorites: 0
- Popularity score: 0