Tags¶

Browse the full Agent Patterns library by topic tag.

Tags are the primary topic-first index for this site. Each topic tag below gets a short intro and anchor pages; the full auto-generated tag listing follows beneath the intros. Tool-specific tags (claude, copilot, cursor, tool-agnostic) carry no prose — see the auto-list at the bottom.

Topic intros¶

context-engineering¶

Patterns for managing what enters the agent's context window: budgets, compression, attention curves, retrieval, and prompt caching. Read this if you're hitting context limits, debugging "lost in the middle" failures, or paying too much for token-heavy prompts. Start with:

agent-design¶

How a single agent is structured: harness, delegation, backpressure, composition, and memory. Read this when picking a harness, designing a long-running loop, or choosing between role-based and task-based delegation. Start with:

multi-agent¶

Orchestration across multiple agents: fan-out, hand-offs, file-based coordination, and sub-agent topology. Read this when one agent isn't enough and the question shifts from "how do I prompt it" to "how do I wire them together". Start with:

memory¶

What an agent remembers across turns, sessions, and projects — versioned vs streaming, episodic vs reinforced, and the failure modes when memory leaks into wrong contexts. Read this when designing persistence or chasing memory-induced drift bugs. Start with:

code-review¶

Agentic code review patterns: how agents read diffs, when committee or tiered reviews pay off, and how to wire review verdicts back into iteration. Read this when integrating agents into PRs or designing a review-then-implement loop. Start with:

instructions¶

System prompts, CLAUDE.md / AGENTS.md files, polarity rules, and the compliance ceiling that limits how many rules an agent will honour. Read this when writing or pruning instruction files, deciding what's primacy-critical, or debugging non-compliance. Start with:

workflows¶

End-to-end agent-assisted development loops: planning, eval-driven iteration, CI integration, onboarding, and team collaboration. Read this when fitting agents into your existing process rather than starting from a blank page. Start with:

cost-performance¶

Routing, token efficiency, model selection, and tool-call cost control. Read this when token spend is hurting, you're tuning a reasoning/execution model split, or designing tool descriptions to minimize per-call overhead. Start with:

testing-verification¶

General testing and verification discipline for agent-generated work: TDD, incremental checks, verification ledgers, and pre-completion gates. Read this for the broad "did the agent do the right thing" patterns — see code-review and evals for the narrower sub-domains. Start with:

evals¶

Evaluation design specifically: pass@k, LLM-as-judge, golden query pairs, and turning incidents into regression cases. Read this when measuring agent quality with structure, building regression suites, or scoring rollouts before they ship. Start with:

security¶

Prompt-injection defense, blast-radius containment, secrets handling, sandboxing, and the lethal-trifecta posture. Read this before granting agents write capability, ingesting untrusted content, or wiring an MCP server. Start with:

observability¶

Tracing, loop detection, transcript analysis, OTel for agents, and harness debugging. Read this when agents fail silently, loops run wild, or you need to diagnose what an agent actually did versus what it claims. Start with:

human-factors¶

The human side of agent-driven work: cognitive load, attention management across parallel sessions, skill atrophy, adoption, and team dynamics. Read this when scaling agents to a team, fighting AI fatigue, or designing supervision rituals. Start with:

github-actions¶

GitHub Actions-specific integration for agent workflows: CI-driven triage, closed-loop remediation, headless runs, and one-click auto-fix. Read this when wiring agents into Actions or designing the CI-side automation that triggers and consumes agent output. Start with:

skills¶

SKILL.md and skill-design patterns: how a named, versioned skill is authored, scoped, and invoked. Spans agent-design and tool-engineering. Read this when designing a CLI-first skill, packaging behavior as an importable module, or gating skill invocation. Start with:

automation¶

Non-interactive, headless agent runs and CI/CD integrations with no human in the loop — scheduled routines, programmatic dispatch, and background pipelines. Read this when wiring agents into CI, cron, or webhook triggers rather than an interactive session. Distinct from workflows, which covers human-driven development loops. Start with:

Tags¶

Topic intros¶

context-engineering¶

agent-design¶

multi-agent¶

memory¶

code-review¶

instructions¶

workflows¶

cost-performance¶

testing-verification¶

evals¶

security¶

observability¶

human-factors¶

github-actions¶

skills¶

automation¶

All tags¶

agent-design¶

anti-pattern¶

arxiv¶

automation¶

claude¶

code-generation¶

code-review¶

context-engineering¶

copilot¶

cost-performance¶

cursor¶

evals¶

fallacies¶

frameworks¶

geo¶

github-actions¶

harness-engineering¶

human-factors¶

index¶

instructions¶

long-form¶

loop-engineering¶

mcp¶

memory¶

multi-agent¶

observability¶

pattern¶

rag¶

reliability¶

sandboxing¶

security¶

skills¶

source:opendev-paper¶

standards¶

supply-chain¶

technique¶

testing-verification¶

token-engineering¶

tool-agnostic¶

tool-engineering¶

training¶

workflows¶