← ClaudeAtlas

graphsmithlisted

Build and fix reliable multi-agent AI workflows for any skill level, from first-time builders to senior engineers. ALWAYS use this skill when the user asks to "build an agent", "automate" a task or workflow, wants "AI to do X for me", describes a multi-step job, mentions agents, bots, pipelines, swarms, or orchestration, or has an automation that "keeps breaking", "forgets where it left off", "loops forever", "duplicates work", or "makes stuff up". Also use to review, harden, or migrate an agent loop to a multi-agent architecture, or when multiple agents or sessions must build in parallel without conflicts ("agents keep overwriting each other", "coordinate parallel agents"). Trigger even if the user never says "agent". Includes an executable scaffolder, architecture linter, crash-recovery test harness, and multi-agent coordination rules, plus KnoSky grounding (auto-updated) so answers about existing code carry citations instead of guesses.
SathiaAI/graphsmith · ★ 2 · AI & Automation · score 68
Install: claude install-skill SathiaAI/graphsmith
# GraphSmith Turn a goal into a reliable multi-agent workflow. One skill, two modes — detect which user you're serving, then apply the same engineering discipline in different voices. **The discipline (never negotiable, in either mode):** 1. A plain, deterministic **manager** controls flow; LLMs work only inside worker steps. Control returns to the manager after every step. 2. **Save after every step** — a crashed run resumes, never restarts. 3. **Every step is safe to re-run** — assume it WILL be retried. 4. **No clocks or randomness in routing** — same inputs, same path. 5. **Minimal handoffs** — workers get only what they need. 6. **Grounded claims** — anything said about existing code cites KnoSky. 7. **One log line per step** — run ID, step, status, duration. --- ## Mode detection (do this first, silently) Score the context. Engineer signals: user mentions frameworks, retries, state, idempotency, CI, reviews existing agent code, or the repo has tests/CI configs. Builder signals: outcome-focused language ("I want it to email my leads"), no code vocabulary, greenfield, or platform-builder context (Lovable/Replit-style). - **Builder mode** → plain English only. Banned words with the user: graph, DAG, node, edge, topology, idempotent, checkpoint, orchestrator, deterministic. Use: blueprint, worker, handoff, save point, manager, "safe to re-run." - **Engineer mode** → full vocabulary, terse output, lead with tools and diffs, skip explanations they didn't ask for. Never