45 docs tagged with "agent-engineering"

Audit an Agent Skills loadout with specification checks, catalog budgets, overlap detection, near-miss prompts, and real activation records.

Write useful AGENTS.md files with correct Codex discovery and precedence, root and nested scopes, practical templates, and maintenance checks.

Map AI Agent engineering roles to interview practice, project evidence, system skills, and a concrete job-preparation path.

Use traces to separate tool, turn, and task failures, test root-cause hypotheses, and turn confirmed failures into eval cases and regression gates.

Build one verifiable AI Agent portfolio project with architecture, tasks, traces, evals, security boundaries, failure logs, and deep-dive evidence.

Practice AI Agent system design through evolving constraints across context, tools, state, loops, graphs, permissions, and verification.

Build an evidence-based AI code review workflow that checks intent, runtime behavior, tests, risk boundaries, and the latest diff before merge.

Use Claude Code prompt-audit to check CLAUDE.md and project instructions, verify stale commands and paths, review edits, and confirm loading in a fresh session.

Build a trusted A2UI Host that validates messages, admits known components, stages batches, and blocks unapproved actions before side effects.

Expose a webpage action with WebMCP, connect it through Playwright MCP, and verify tool results against the page DOM with a pinned browser setup.

Evaluate whether agents use tool results with jevals and Kev. Add human labels, compare threshold tradeoffs, and apply a chosen threshold to held-out traces.

Learn when to use Claude Code Skills, Hooks, or MCP by separating reusable procedures, lifecycle automation, and external tool connections.

Understand Claude Code token usage, /usage fields, active context, prompt cache reads and writes, session cost, and practical ways to diagnose growth.

Build a fair 3–5-task pilot to compare Coding Agents on your repository with fixed commits, tests, permissions, budgets, repeated trials, and review burden.

Learn what a Coding Agent should verify in a real browser before trusting a UI patch: layout, network, console, session state, and evidence limits.

Turn a preserved Coding Agent trace into a local JSONL dataset, deterministic graders, machine-readable reports, and a CI quality gate.

Understand how a coding agent harness controls repository context, tools, orchestration, sandbox boundaries, memory, verification, and agent selection.

Learn how to separate coding-agent memory into durable project rules, active task state, retrievable history, and disposable working context.

Build a local Coding Agent trace audit that preserves tool failures, retries, Token usage, handoffs, and explicit telemetry gaps.

Understand coding agent sandbox boundaries, workspace-write risks, host trust, command allowlists, privileged daemons, and practical isolation checks.

Run a reproducible Copilot CLI 1.0.94 lab to compare local sandbox policies, offline mode, file access, and tool-network access.

Build recalculable DeepSeek API cost records from cache-hit input, cache-miss input, output, retries, and human review time.

Set up DeepSeek V4-Flash in Codex CLI with a model catalog, named profile, API key, verification, rollback, and Responses API compatibility limits.

Find code by behavior with Jevgrep, inspect source evidence, and give your coding agent a focused implementation task.

Run a coding task with the OpenAI Agents API, continue it in a persistent session, inspect execution output, download artifacts, and recover after a stream disconnects.

Install Hermes Agent, connect a model provider, and read a local Python project with file references. Verify its predictions, resume sessions, and troubleshoot setup.

Design a Coding Agent Graph with explicit dependencies, serial and parallel execution, typed signals, merge rules, escalation, and governor authority.

Review AI-generated tests against requirements, inspect assertions and boundaries, and use meaningful mutations to identify missing tests.

Use oasdiff to detect breaking OpenAPI changes, identify affected client requests, and give a coding agent a focused repair task.

Run parallel Coding Agents with isolated worktrees, runtime state, evidence, and merge authority so concurrency does not become integration chaos.

Turn a small feature request into acceptance criteria for a coding agent, with boundary cases, a copyable task brief, and a verification record.

Verify AI-generated Go code with a reproducible Go workflow covering build, tests, vet, fuzzing, govulncheck, and human review.

Connect a generated OpenAPI TypeScript client to project authentication, timeouts, and error handling, then give an AI agent concrete acceptance checks.

Install Kimi Code CLI, migrate safely from the legacy Kimi CLI, verify configuration, understand permission modes, and run a careful first task.

Review dependent coding-agent changes with a local Git and Python lab: isolate layer diffs, catch semantic regressions after rebase, and repair squash-merge baselines.

Build a bounded Coding Agent Loop with an explicit local aim, evaluator, budget, stopping condition, authority scope, and escalation path.

Use a practical MCP server security checklist to verify provenance, OAuth scopes, token handling, tool side effects, data access, logging, and revocation.

Design MCP Tools that help Agents choose the right action through task-shaped boundaries, schemas, honest annotations, compact results, and actionable errors.

Upgrade Jest 29 to 30 with a coding agent: migrate matcher aliases and CLI flags while preserving assertion intent, using NestJS as a reference.

Run a scoped audit with Cloudflare Security Audit Skill, read confirmed and unresolved findings, and turn evidence into a focused repair task.

Run Kev-0.8B locally on Apple Silicon, send a Choice request for support ticket routing, and understand model metadata, probabilities, and confidence.

Learn a practical GitHub Spec Kit workflow from constitution and specification through plan, tasks, implementation, and human review gates.

Compare local and CI test logs, source, dependencies, timezones, and event order to guide an AI agent toward a verifiable fix.

Migrate TypeScript MCP SDK v2 to protocol 2026-07-28 with explicit negotiation, stateless HTTP handling, state relocation, and legacy client checks.

Plan a phased PostgreSQL 18 required-field migration with an AI coding agent, covering existing rows, older application versions, validation, and rollback.