10 Skills of the Day

Use claude agents --json to build session dashboards. Claude Code v2.1.145 outputs all live agent sessions as structured JSON with status, model, elapsed time, and parent relationships. Pipe it into a tmux status bar widget or session picker script for switching between background agents in real time.

Check your vendors against the SaaStr API Report Card before renewing contracts. 144 B2B APIs are now graded on agent readiness at saastr.com. If your CRM or analytics platform scores below 60, plan your migration now before agent workflows hit a wall.

Add a cross-encoder reranker to your RAG pipeline for 18-42% precision gains. Hugging Face's new Ettin Reranker family is open-source and drops in between your retriever and LLM. Minimal latency cost, measurable quality improvement on retrieval-heavy workloads.

Install code-review-graph for 6.8x token savings on Claude Code sessions. Set up the MCP server, let it build a Tree-sitter AST knowledge graph (~10 seconds for 500 files), and Claude reads only structurally relevant code. 49x reduction on daily tasks.

Run agent workloads through a cost calculator before committing to cloud. Dell's deskside AI claims 87% cost reduction with 3-month break-even. For repetitive agent tasks like code review, testing, or CI analysis, local inference on a workstation may already be cheaper than API calls.

Audit your cloud account for single-point-of-failure risks. After the Railway outage: keep your domain registrar, DNS, and status page on a separate provider and separate billing account. Build a static fallback page on a different cloud. Account-level suspension is automated and can hit anyone.

Implement URL allowlist enforcement on any agent processing external input. The Salesforce ForcedLeak vulnerability (CVSS 9.4) was indirect prompt injection through a web form that tricked an agent into exfiltrating data. If your agents process user-submitted content, constrain which external URLs they can reach.

Try Gemini 3.5 Flash for batch agent workloads at $1.50/M input tokens. It outperforms the previous Pro model at 4x the speed with a 1M-token context window. For monitoring pipelines, batch analysis, or any token-heavy job, run the numbers against your current provider.

Add persistent memory to your Claude Code workflow with AgentMemory. rohitg00/agentmemory provides 95.2% recall using hybrid BM25 + Vector + Graph search with 53 MCP tools, at roughly $10/year. Your sessions start compounding context instead of resetting every conversation.

Use GitHub's new spec-kit to generate implementation plans from specifications. The official toolkit ships /speckit.specify and /speckit.implement commands that work across 30+ coding agents. Write the spec once, and the agent handles decomposition and implementation. Spec-driven development with agent execution is now a first-party GitHub workflow.