Help us improve
Share bugs, ideas, or general feedback.
From orchestrator-core
Execute an orchestrator-native plan from .claude/plans/, dispatching agents per the plan's agent map and enforcing the 5 invariants throughout.
npx claudepluginhub rafa1off/orchestrator --plugin orchestrator-coreHow this skill is triggered — by the user, by Claude, or both
Slash command
/orchestrator-core:orchestrator-executeWhen to use
Use after orchestrator-plan has produced a plan document. Handles TaskCreate, agent dispatch, the verification loop, and the final summary.
The summary Claude sees in its skill listing — used to decide when to auto-load this skill
**Announce at start:** "Using orchestrator-execute to implement the plan."
Guides technical evaluation of code review feedback: read fully, restate for understanding, verify against codebase, respond with reasoning or pushback before implementing.
Share bugs, ideas, or general feedback.
Announce at start: "Using orchestrator-execute to implement the plan."
.claude/plans/Read the plan's Stages list. Create one task per stage using the exact stage description from the plan:
TaskCreate("Stage N — [agent]: [exact description from plan]")
For Stage Na — checker and Stage Nb — reviewer entries, create two separate tasks — one for each:
TaskCreate("Stage Na — checker: [exact scope from plan]")
TaskCreate("Stage Nb — reviewer: [exact scope from plan]")
If the plan has 3+ stages, also create .claude/pipeline/progress.md:
# Pipeline Progress — [task name from plan Goal]
## Status
Step: Stage 1
## Completed
(none yet)
## Pending
- [ ] Stage 1 — [agent]: [exact description from plan]
- [ ] Stage 2 — [agent]: [exact description from plan]
- [ ] Stage 3a — checker: [exact scope from plan]
- [ ] Stage 3b — reviewer: [exact scope from plan]
...
## Key Decisions
(none yet)
## Open Issues
(none yet)
Work through stages in plan order. The 5 invariants are non-negotiable — they override anything in the plan.
TaskUpdate(id, status="in_progress")
Agent(reader, "[instruction from plan stage]")
TaskUpdate(id, status="completed")
Update progress.md: move stage to Completed with a one-line note on what was found.
TaskUpdate(id, status="in_progress")
Agent(researcher, "[instruction from plan stage]")
TaskUpdate(id, status="completed")
TaskUpdate(id, status="in_progress")
Agent(thinker, """
[context block]
[question or decision from plan stage]
""")
TaskUpdate(id, status="completed")
Serialize if file sets overlap with another active writer stage.
TaskUpdate(id, status="in_progress")
Agent(writer, """
## Context
[reader output relevant to this stage]
## Task
[instruction from plan stage]
## Files to modify
[exact paths from plan]
""")
TaskUpdate(id, status="completed")
Always dispatched together in the same message turn.
rm -f .claude/pipeline/checker-findings.json .claude/pipeline/reviewer-findings.json
TaskUpdate(checker_id, status="in_progress")
TaskUpdate(reviewer_id, status="in_progress")
Agent(checker, "Run scoped checks. Modified files: [list from writer output]")
Agent(reviewer, "[task context and summary of what writer changed]")
After both complete:
cat .claude/pipeline/checker-findings.json
cat .claude/pipeline/reviewer-findings.json
If both PASS/APPROVED:
TaskUpdate(checker_id, status="completed")
TaskUpdate(reviewer_id, status="completed")
Continue to next stage.
If either has findings: Send merged batch to writer:
## Batch Fixes Required
### Checker errors
[from checker-findings.json, or "none"]
### Reviewer issues
[from reviewer-findings.json, or "none"]
Re-verify (back to rm -f step). After 2 full rounds with remaining findings: stop, surface all findings to user, ask for direction. Never continue silently.
TaskUpdate(id, status="in_progress")
Agent(tester, """
Task: [description from plan]
Changed files: [list]
Write and run tests for: [what the plan specifies]
""")
TaskUpdate(id, status="completed")
TaskUpdate(id, status="in_progress")
Agent(documenter, """
Changed public surface: [what changed]
Files modified: [list]
Update: [doc targets specified in plan]
""")
TaskUpdate(id, status="completed")
## Done
**Task:** [original task from plan Goal]
**Status:** ✅ Done / ⚠️ Escalated
**Changes:**
- [file] — [what changed]
**Tests:** [N new, all passing / N failing / not applicable]
**Docs:** [updated / not needed]
**Review:** APPROVED / [open issues]