Developer Productivity & Operational Enablement

Knowledge of

• The CLAUDE.md hierarchy: enterprise/system, user (~/.claude), project (./CLAUDE.md), and subdirectory levels, auto-loaded into context, with the project-level CLAUDE.md committed to the repo so the whole team shares the same conventions
• settings.json as the deterministic control surface for tool allow/deny lists, hooks, environment variables, model selection, and connected MCP servers -- distinct from CLAUDE.md's memory/instructions role
• Shared MCP servers as build-once, reuse-across-the-team configuration for common integrations (internal APIs, data sources)
• Team-wide Skills distribution via repos, plugins, or enterprise managed settings so procedures and know-how are shared rather than reinvented per developer

Skills in

• Diagnosing a misconfiguration where permissions or tool allow/deny rules were mistakenly placed in CLAUDE.md instead of settings.json
• Standardizing team-wide Claude Code behavior by committing a project-level CLAUDE.md to the repo rather than relying on each developer's personal configuration
• Configuring a shared MCP server once for a common integration instead of having every developer configure the same integration individually
• Distributing reusable Skills via repos, plugins, or enterprise managed settings so know-how is shared team-wide

Knowledge of

• The gather-context to plan to act to verify loop as Claude Code's core best-practice workflow, keeping context lean to avoid context rot
• The Agent SDK enabling teams to build custom internal agents and automate workflows beyond the interactive tool (CI/CD checks, codebase modernization, repetitive engineering tasks)
• Headless/non-interactive modes for running Claude Code in scripts and CI pipelines
• Custom slash commands and automation (e.g., GitHub integration) for standardizing common team tasks

Skills in

• Structuring an AI-assisted workflow around gather-context, plan, act, verify rather than skipping the verify step for speed
• Keeping context lean during iterative work to avoid context rot degrading output quality
• Using the Agent SDK to build custom internal agents for CI/CD automation and repetitive engineering tasks
• Running Claude Code in headless/non-interactive mode to extend AI assistance into CI pipelines
• Standardizing common team tasks with custom slash commands and GitHub integration automation

Knowledge of

• Isolating a failure to its originating layer -- transport (HTTP/auth errors), integration/parsing code, retrieval (stale/irrelevant chunks), or model output (well-formed but wrong) -- before applying a fix
• Using traces and logs (request/response pairs, tool calls, retrieval hits, stop_reason, token usage) to walk back to the earliest deviation rather than just the final symptom
• Checking operational basics: truncated output (stop_reason: max_tokens), model version mismatches, and context bloat late in a session
• Feeding confirmed fixes back into evals and monitoring so recurring issues are caught automatically next time

Skills in

• Localizing a production issue to the transport, integration/parsing, retrieval, or model-output layer before changing any code or prompts
• Tracing request/response pairs, tool calls, retrieval hits, stop_reason, and token usage back to the earliest point of deviation
• Diagnosing truncated output by checking for stop_reason: max_tokens and raising the max_tokens parameter rather than assuming a quality or security problem
• Recognizing model version mismatches and context bloat as operational root causes distinct from prompt or code defects
• Feeding confirmed fixes back into eval suites and monitoring dashboards to prevent the same issue from recurring