Read a plan document, decompose it into steps, design a per-step agent chain from the ECC catalogue, and emit ready-to-paste /orchestrate custom prompts. Generative only — never invokes /orchestrate itself. Use when the user has a multi-step plan and wants to drive it through orchestrate without composing chains by hand.
The plan-orchestrate skill bridges a plan document to the /orchestrate custom command by reading the plan, decomposing it into steps, designing a per-step agent chain from the ECC (Everything Claude Code) catalogue, and emitting one ready-to-paste /orchestrate custom invocation per step. It is generative only — it never executes /orchestrate itself; the user pastes each emitted line when ready. It activates when a user has a multi-step plan document (PRD, RFC, or implementation plan) and wants to drive it through orchestrate without composing agent chains by hand, and it skips one-off ad-hoc steps or unreadable/empty plans.
Inputs follow the shape <plan-doc-path> [--lang=python|typescript|go|rust|cpp|java|kotlin|flutter|auto] [--scope=all|step:<n>|range:<a>-<b>] [--dry-run], where --lang defaults to auto (detected from the project), --scope defaults to all, and --dry-run prints the decomposition and chain rationale without emitting final prompts. The authoritative output form is {ORCH_CMD} custom "<agent1>,<agent2>,..." "<task description>", a sequential comma-separated chain where each agent's HANDOFF feeds the next. The skill detects the ECC install form once and freezes it: a plugin install (marketplaces/ecc present under the Claude home) yields /ecc:orchestrate with ecc:<name> agents, while a legacy bare install yields /orchestrate with bare names — the two must stay in sync and never mix in one output. Phase 0 also resolves language via project markers (pyproject.toml, package.json, go.mod, Cargo.toml, and others), applies a polyglot tie-break by counting source files, and detects a PyTorch sub-profile. Later phases decompose steps by numbering, tables, or headings, then tag each step and pick a default chain from a trigger-word table.
The skill draws only from a fixed agent catalogue — including planner, architect, tdd-guide, code-reviewer, security-reviewer, refactor-cleaner, doc-updater, docs-lookup, e2e-runner, database-reviewer, language-specific build-error-resolvers and reviewers, and others — and warns that a misspelled or invented agent name, or invented flags like --mode or --gate, will fail /orchestrate. It targets ECC users decomposing structured plans into orchestrated agent chains. Key requirements are a readable, non-empty plan document and a correctly detected ECC install form; the lang=unknown sentinel is not an agent name and is resolved into code-reviewer or build-error-resolver at chain-composition time.
No. The skill is generative only — it never executes /orchestrate. It emits one ready-to-paste invocation per step, and you paste each line when ready.
It takes <plan-doc-path> plus optional flags: --lang (python, typescript, go, rust, cpp, java, kotlin, flutter, or auto), --scope (all, step:<n>, or range:<a>-<b>), and --dry-run to print only the decomposition and chain rationale.
It detects the install form once and freezes it. A plugin install uses /ecc:orchestrate with ecc:<name> agents; a legacy bare install uses /orchestrate with bare names. The slash command and agent names must stay in sync, and the two forms are never mixed in the same output.
No. Agent names must come from the skill's fixed catalogue (planner, architect, tdd-guide, code-reviewer, and others). A misspelled agent name fails /orchestrate, so names are cross-checked against the list before emitting.
No. No --mode, --gate, or --agents=... flags exist — the skill never invents them. The command is always {ORCH_CMD} custom "<agents>" "<task description>".
Quick Setup:
.claude/skills/Repository
affaan-m/everything-claude-code