agentsclimarketplace

Orchestrator

Skill shihwesley/shihwesley-plugins/skills/orchestrator

Automated plan execution pipeline. Use when running /orchestrate on an /interactive-planning output. Handles plan ingestion, review, skill matching, tech research, user gating, and phased execution with worktree isolation, testing, code review, and incremental commits.From its SKILL.md

Install
npx -y skills add shihwesley/shihwesley-plugins --skill orchestrator

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

4.8 KB, as published. Nobody here has run it

Orchestrator

Takes /interactive-planning output and executes it through a 6-stage pipeline: ingest, review, skill-match, research, user gate, and phased execution with worktree isolation.

When to Use

  • After /interactive-planning produces a plan directory with task_plan.md or manifest.md
  • When you need automated, isolated execution of a multi-phase plan
  • When resuming an interrupted orchestration (--resume)

Pipeline Stages

  1. INGEST -- read plan files (task_plan.md or manifest.md + specs), discover project docs (CLAUDE.md, codebase map)
  2. REVIEW -- agent checks phase sizing, gaps, clarity, dependencies, confidence (skipped on --resume)
  3. MATCH -- resolve skills per phase via registry lookup, AgentReverse fallback, web search fallback, or general-purpose default
  4. RESEARCH -- fetch official docs for unfamiliar tech (Context7, knowledge store, web), produce cheat sheets
  5. GATE -- present full execution plan to user for approval
  6. EXECUTE -- for each phase sequentially, launch a phase-runner subagent that handles worktree creation, agent dispatch, testing, code review, commit, and merge

Dispatch Modes

The orchestrator picks classic or team mode per-phase based on task count, shared files, and dependencies:

ConditionMode
1-2 tasksclassic
3+ tasks, all independentclassic
4+ tasks with shared files or depsteam
Low/medium confidence, 3+ tasksteam
--team flagteam (forced)
--no-team flagclassic (forced)

Flags

FlagEffect
--dry-runRun stages 1-4, show plan at stage 5, stop
--resumeSkip completed phases, skip stage 2 review
--max-parallel NMax agents per batch (default: 3)
--model sonnet|opus|haikuModel for dispatched agents
--phase NExecute only phase N
--teamForce Agent Teams mode
--no-teamForce classic Task dispatch

Component Files

All component specifications live in references/:

FileRole
references/plan-ingester.mdParse plan files, handle spec-driven and task-based modes
references/plan-reviewer.mdReview and restructure plans (sizing, gaps, clarity)
references/skill-matcher.md3-step skill resolution cascade
references/tech-researcher.mdFetch docs, produce cheat sheets, manage knowledge store
references/skill-registry.jsonKeyword-to-skill/agent mappings
references/worktree-manager.mdWorktree lifecycle per phase
references/agent-dispatcher.mdDual-mode dispatch (team or classic)
references/team-lead.mdAgent Teams lead coordination protocol
references/phase-runner.mdSelf-contained protocol for one phase
references/phase-finisher.mdTest, review, commit chain (runs inside phase-runner)

Context Window Design

The main orchestrator stays lean (~800 tokens per phase) by delegating all heavy work to phase-runner subagents. Each phase-runner gets its own context window and returns a ~20-line structured result. A 5-phase plan uses roughly 4k tokens for execution.

Examples

# Standard execution
/orchestrate ./plans/my-feature

# Preview without executing
/orchestrate --dry-run ./plans/my-feature

# Resume after interruption
/orchestrate --resume ./plans/my-feature

# Execute specific phase with limited parallelism
/orchestrate --phase 2 --max-parallel 2 ./plans/my-feature

# Force team mode
/orchestrate --team ./plans/my-feature

Common Issues

  • Plan file not found: The orchestrator resolves PLAN_DIR to an absolute path via realpath. Make sure the directory exists and contains task_plan.md or manifest.md.
  • No skills matched: Falls back to general-purpose agent type with a warning. Add keywords to your plan's task descriptions to improve matching.
  • Phase-runner fails after retries: You get four options: fix manually, skip the phase, abort, or retry from scratch. The phase-runner's error message usually points to the root cause.
  • Context window fills up: The phase-runner architecture prevents this by design. If the main orchestrator still fills up, it's likely from stages 1-4 reading too many files. Use --resume to skip stages on retry.
  • Merge conflicts: The orchestrator pauses and asks you to resolve manually. It won't force-merge.

Keep looking

Skills are one crate of 609,689. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.