Use when you have an implementation plan and want to execute it. Analyzes the plan's task graph, determines the best execution strategy (parallel via team-dev or serial via sdd), and orchestrates the full build with quality gates. The single entry point for plan execution — you don't need to choose between team-dev and sdd yourself.
From casaflownpx claudepluginhub casaperks/casaflow --plugin casaflowThis skill uses the workspace's default tool permissions.
Designs and optimizes AI agent action spaces, tool definitions, observation formats, error recovery, and context for higher task completion rates.
Enables AI agents to execute x402 payments with per-task budgets, spending controls, and non-custodial wallets via MCP tools. Use when agents pay for APIs, services, or other agents.
Compares coding agents like Claude Code and Aider on custom YAML-defined codebase tasks using git worktrees, measuring pass rate, cost, time, and consistency.
PURPOSE: Execute an implementation plan. Analyzes the task graph, picks the right execution strategy (parallel or serial), and runs it with quality gates. You hand it a plan — it builds the thing.
CONFIGURATION: Reads jig.config.md for parallel-threshold, default-strategy, and teammate-mode.
brainstorm → plan produces an implementation plandocs/plans/*.md and want to execute itteam-dev or sddYou don't need to know the difference between parallel and serial execution. This skill figures it out.
flowchart TD
find[Find the implementation plan] --> read[Read plan and extract tasks]
read --> analyze{Analyze task graph}
analyze --> count{Independent tasks<br/>≥ parallel threshold?}
count -->|yes| files{Tasks touch<br/>different files?}
count -->|no| serial[Execute via sdd<br/>serial with review gates]
files -->|yes| teams{Agent teams<br/>available?}
files -->|no| serial
teams -->|yes| parallel[Execute via team-dev<br/>parallel with quality gates]
teams -->|no| serial
parallel --> done[All tasks complete]
serial --> done
done --> verify[Run verification]
verify --> finish[Hand off to finish]
Look for the implementation plan in this order:
docs/plans/ for the newest *-plan.md fileplan first.Read the plan file. Extract:
blockedBy relationships)Read jig.config.md for:
parallel-threshold (default: 3) — minimum independent tasks for paralleldefault-strategy (default: team-dev) — preference when conditions are metteammate-mode (default: tmux) — how parallel agents appearDecision logic:
| Condition | Strategy | Why |
|---|---|---|
| < threshold independent tasks | sdd (serial) | Not enough parallelism to justify team overhead |
| ≥ threshold but tasks share files | sdd (serial) | File conflicts make parallel unsafe |
| ≥ threshold, different files, teams available | team-dev (parallel) | Full parallel execution with quality gates |
| ≥ threshold, different files, teams NOT available | sdd (serial) | Fallback — suggest enabling agent teams |
| Only 1-2 tasks total | Direct execution | No orchestrator needed — just do it |
Announce the decision:
"This plan has N tasks. M are independent and touch different files. Using team-dev (parallel execution with quality gates)."
or
"This plan has N tasks but they're tightly coupled (shared files / sequential dependencies). Using sdd (serial execution with review gates)."
or
"This plan has 2 tasks. Executing directly — no orchestrator needed."
If parallel (team-dev):
Invoke the team-dev skill with the full plan. It handles:
If serial (sdd):
Invoke the sdd skill with the full plan. It handles:
If direct (1-2 tasks): Execute tasks inline in the current session:
After all tasks complete (regardless of strategy):
Run verification — use verify to confirm everything works:
Hand off to finish — which presents options:
pr-create)If the plan references a PRD (> **PRD:** docs/plans/...), the spec reviewers in both team-dev and sdd will automatically load the PRD's acceptance checklist and verify each [ ] item against the implementation. You don't need to do anything extra — the link in the plan header is enough.
| Situation | Action |
|---|---|
| No plan found | Ask the user. Suggest /casaflow:plan to create one. |
| Plan has no tasks | The plan is incomplete. Ask the user to review it. |
| Agent teams not available for parallel | Fall back to serial. Note: "Agent teams aren't enabled. Using serial execution. To enable: add CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS: "1" to settings." |
| Task fails review repeatedly | The strategy handles this (team-dev sends feedback, sdd re-reviews). If 3+ review cycles fail, escalate to the user. |
| Build/tests fail after all tasks | Don't hand off to finish. Fix the issues first, re-verify. |
| Scope changed mid-build | Pause. Ask: "The implementation is diverging from the plan. Should we update the plan or continue?" |
Called by:
kickoff during the EXECUTE stage/casaflow:buildInvokes:
team-dev — for parallel executionsdd — for serial executionverify — after all tasks completefinish — to complete the branchRelated:
plan — creates the plan that this skill executesprd — produces acceptance checklists that spec reviewers verify during executionkickoff — the full pipeline orchestrator that routes here at the EXECUTE stage/casaflow:build "Here's a plan, build it"
│
┌─────────┴──────────┐
│ Analyze task graph │
└─────────┬──────────┘
│
┌───────────────┼───────────────┐
│ │ │
≥ threshold < threshold 1-2 tasks
+ diff files or shared files
│ │ │
team-dev sdd direct
(parallel) (serial) (inline)
│ │ │
└───────────────┼───────────────┘
│
┌─────────┴──────────┐
│ verify → finish │
└────────────────────┘