Skip to content

cogni-dao/cogni / .claude/skills/agent-infrastructure-scorecard

agent-infrastructure-scorecard

Verified

Living status scorecard for Cogni agent infrastructure — InProc↔LangGraph-Server alignment, built-vs-designed inventory, and the agent/langgraph doc DRY/drift/consolidation state. The stable mental model lives in agent-infrastructure-expert; this is the dated state that moves as graphs, executors, and evals ship. Use to check "what's actually built vs paper", "where do InProc and Server diverge", "which agent specs are stale/duplicated/safe-to-delete". Update THIS, not the expert skill.

Agent Infrastructure Scorecard

Living state under agent-infrastructure-expert. The expert skill holds the stable mental model + operating rules; this scorecard holds dated status that moves as things ship. When something changes, refine this file — the skill stays constant.

Last verified: 2026-06-09

InProc ↔ LangGraph Server Alignment

https://node-template.cognidao.org/knowledge/graph-execution-invariants north-star: "InProc must model as closely to LangGraph Server's I/O as possible." Where they stand:

DimensionInProc (live, P0)LangGraph Server (designed)Aligned?
GraphExecutorPort✅✅🟢 same port
providerId / graphIdlanggraph / langgraph:<name>langgraph / langgraph:<name>🟢 backend swaps via env, not id
Output vocabularyai-core AiEventai-core AiEvent🟢 nothing vendor-specific crosses
Tool eventsfull (tool_call_*)P0: text/usage/done/error only🟡 InProc richer
stateKey / threadsignored — no persistencerequired; UUIDv5 tenant-scoped checkpoints🔴 divergent
Resume / time-travelnone (graph loss = full re-run)native checkpointer (Redis)🔴 divergent
Billing pathstream usage_report → commitUsageFact()async reconciliation via LiteLLM end_user/spend-logs🔴 different mechanism
LLM routingCogniCompletionAdapter (ALS completionFn)LiteLLM proxy, per-user virtual key🟡 different
Deploymentonly live path — bundled in app imagenot deployed (no catalog target, no LANGGRAPH_SERVER_URL in any env)🔴 Server is paper

Takeaway: InProc is production; Server is a spec + compose file with no running instance. They diverge most exactly where durability lives (threads/resume/billing) — the same seam as the synchronous-HTTP execution boundary. Never claim Server parity in a design without confirming it's actually deployed.

Built vs Designed-Only

CapabilityStatus
InProc execution, billing, observability, credit preflight🟢 built
Temporal orchestration + HTTP-delegated graph runs🟢 built
Node-sovereign graph packages (@cogni/<node>-graphs)🟢 built
LangGraph Server executor🔴 spec + compose only, not deployed
Evals — datasets, LLM-judge, CI gate, canary gate🔴 0/8; evals/ dir does not exist; nothing gates promotion (proj.ai-evals-pipeline.md, task.0286)
UI graph picker🟡 AVAILABLE_GRAPHS hardcoded in ChatComposerExtras.tsx, not from /api/v1/ai/agents

DRY, Drift & Consolidation — agent/langgraph cluster

Evidence-backed. CICD docs are out of scope and on HOLD until the pipeline is green — no edits to https://cognidao.org/knowledge/operator-pipeline-contract / cd-pipeline-*.

#FindingEvidenceFix
1Invariant duplication; https://node-template.cognidao.org/knowledge/graph-execution-invariants is SSOT yet others restateGraphExecutorPort+AiEvent in all 4 of graph-execution / langgraph-patterns / langgraph-server / unified-graph-launch; PACKAGES_NO_SRC_IMPORTS ×3; NO_LANGCHAIN_IN_SRC ×2each restated invariant → one-line link to graph-execution
2Speculative executor specs for unbuilt pathsclaude-sdk-adapter (not implemented), n8n-adapter (P2), multi-provider-llm (future), completions-api (proposed) — all draftcollapse to one "Future Executors" stub under graph-execution
3Path driftapps/operator (real nodes/<node>/app) in unified-graph-launch, ai-pipeline-e2e, multi-provider-llm; ../LANGGRAPH_SERVER.md (real https://node-template.cognidao.org/knowledge/langgraph-server-adapter) in langgraph-patternsmechanical rewrite
4agent-development.md operator-centric + missing ship/runleads shared Tier 1a; no runtime/CI linklead node-local packages/graphs; add Build→Ship→Run paragraph
5Paradigm doc masquerading as guideagent-design.md = KPIs/paradigm, not a runbook→ operator knowledge Dolt. Rule: guide = runbook; paradigm = knowledge entry

Target shape: ~6 core specs (graph-execution = single invariant SSOT, + langgraph-patterns, langgraph-server, temporal-patterns, unified-graph-launch, ai-pipeline-e2e) and 3 guides (agent-development, tools-authoring, langgraph-server). Everything else links up or merges; nothing restates invariants.

Canonical Doc Map (read order + stale flags)

  1. graph-execution.md — authoritative spec. Executor port, decorator stack, routing, catalog, ALS.
  2. langgraph-patterns.md — package boundaries, InProc flow, anti-patterns. 🟡 fix ../LANGGRAPH_SERVER.md link.
  3. unified-graph-launch.md — run topology (Temporal → app → Redis → SSE). 🟡 apps/operator drift.
  4. temporal-patterns.md — durability boundary, webhook→workflow→graph→write.
  5. langgraph-server.md — alternate executor (designed, not live).
  6. ai-pipeline-e2e.md — auth→execution→billing→security E2E. 🟡 apps/operator drift.

Narrower/adjacent: agent-discovery, agent-registry, ai-setup, sandboxed-agents, node-baas-architecture. Speculative (see DRY #2): claude-sdk-adapter, n8n-adapter, multi-provider-llm, completions-api. Authoring tier: agent-development.md 🟡, agent-design.md → knowledge, tools-authoring.md.

writing-skillsobra/superpowersUse when creating new skills, editing existing skills, or verifying skills work before deploymentStars: 322kUpdated Claude CodeCodexVerifiedInstall
xlsxanthropics/skillsUse this skill any time a spreadsheet file is the primary input or output. This means any task where the user wants to: open, read, edit, or fix an existing .xlsx, .xlsm, .xltx, .csv, or .tsv file (e.g., adding columns, computing formulas, formatting, charting, cleaning messy data); create a new spreadsheet from scratch or from other data sources; or convert between tabular file formats. Trigger especially when the user references a spreadsheet file by name or path — even casually (like "the xlsx in my downloads") — and wants something done to it or produced from it. Also trigger for cleaning or restructuring messy tabular data files (malformed rows, misplaced headers, junk data) into proper spreadsheets. The deliverable must be a spreadsheet file. Do NOT trigger when the primary deliverable is a Word document, HTML report, standalone Python script, database pipeline, or Google Sheets API integration, even if tabular data is involved.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
shipping-and-launchaddyosmani/agent-skillsPrepares production launches. Use when preparing to deploy to production, or when asking what needs to be in place before shipping. Use when you need a pre-launch checklist, when setting up monitoring, when planning a staged rollout, or when you need a rollback strategy.Stars: 119kUpdated Claude CodeCodexVerifiedInstall
childhoodthedotmack/claude-memBabysit a shipped release after merge, publication, or deployment. Verify published packages, production behavior, installed plugins, and late CI or user reports until the observation window is clean; use babysit for PR readiness before release.Stars: 118kUpdated Claude CodeCodexVerifiedInstall