object-objectlisted

Development orchestration skill for {{project_name}}. Use for feature development, changes, test hardening, and code-review requests, running them through a multi-agent pipeline. Also use for resuming a previous run and applying review feedback.
Korea-DongheeHan/graph-builder · ★ 0 · AI & Automation · score 70

Install: claude install-skill Korea-DongheeHan/graph-builder

# {{pipeline_name}} — development orchestration A **self-contained orchestration** generated by agent-pipeline-builder. It runs from this directory plus `.claude/agents/{{prefix}}-*.md` alone (the plugin is not required). - Flow (SSOT): `pipeline.yml` — nodes, parallelism, branches, feedback loops - Roles and expertise: `.claude/agents/{{prefix}}-*.md` (model, tools, system prompt) - Task input and verdict criteria: `prompts/*.md` - Execution engine: `scripts/run_graph.py` (state, branching, parallelism, resume) ## Execution modes | Mode | How | Traits | |---|---|---| | **Runner** | `python3 scripts/run_graph.py pipeline.yml` | Deterministic, state-saving, resumable. Each node is an isolated claude session. Progress via console log. Zero orchestration cost | | **Session** | Claude executes nodes with the Agent tool per `references/session-mode.md` | Live per-node subagent view and mid-run intervention. No resume; main-session overhead | **The default mode is `settings.mode` in `pipeline.yml`.** Override only when the user asks ("in session mode / let me watch", or "in runner mode"). Unattended, batch, and large runs suit the runner; observation, debugging, and intervention suit the session. ## Procedure (when Claude executes this skill) Call this skill directory **GRAPH_PIPELINE** (`.claude/skills/{{pipeline_name}}`). Run from the project root (agent definitions and build commands resolve from there). 0. **Determine the run context** — check the latest run in `.graph-