The variety of choices we’ve to configure and enrich a coding agent’s context has exploded over the previous few months. Claude Code is main the cost with improvements on this house, however different coding assistants are rapidly following swimsuit. Highly effective context engineering is turning into an enormous a part of the developer expertise of those instruments.
Context engineering is related for all sorts of brokers and LLM utilization after all. My colleague Bharani Subramaniam’s easy definition is: “Context engineering is curating what the mannequin sees so that you just get a greater consequence.”
For coding brokers, there’s an rising set of context engineering approaches and phrases. The inspiration of it are the configuration options supplied by the instruments (e.g. “guidelines”, “abilities”), after which the nitty gritty of half is how we conceptually use these options (“specs”, numerous workflows).
This memo is a primer concerning the present state of context configuration options, utilizing Claude Code for example on the finish.
What’s context in coding brokers?
“The whole lot is context” – nevertheless, these are the principle classes I consider as context configuration in coding brokers.
Reusable Prompts
Virtually all types of AI coding context engineering finally contain a bunch of markdown recordsdata with prompts. I take advantage of “immediate” within the broadest sense right here, prefer it’s 2023: A immediate is textual content that we ship to an LLM to get a response again. To me there are two foremost classes of intentions behind these prompts, I’ll name them:
Directions: Prompts that inform an agent to do one thing, e.g. “Write an E2E take a look at within the following means: …”
Steerage: (aka guidelines, guardrails) Common conventions that the agent ought to comply with, e.g. “At all times write assessments which can be impartial of one another.”
These two classes usually mix into one another, however I’ve nonetheless discovered it helpful to differentiate them.
Context interfaces
I couldn’t actually discover a longtime time period for what I’d name context interfaces: Descriptions for the LLM of the way it can get much more context, ought to it determine to.
Instruments: Constructed-in capabilities like calling bash instructions, looking recordsdata, and so forth.
MCP Servers: Customized applications or scripts that run in your machine (or on a server) and provides the agent entry to knowledge sources and different actions.
Expertise: These latest entrants into coding context engineering are descriptions of further sources, directions, documentation, scripts, and so forth. that the LLM can load on demand when it thinks it’s related for the duty at hand.
The extra of those you configure, the more room they take up within the context. So it’s prudent to assume strategically about what context interfaces are essential for a specific process.
Information in your workspace
Probably the most fundamental and highly effective context interfaces in coding brokers are file studying and looking, to grasp your
If and when: Who decides to load context?
LLM: Permitting the LLM to determine when to load context is a prerequisite for working brokers in an unsupervised means. However there at all times stays some uncertainty (dare I say non-determinism) if the LLM will really load the context once we would count on it to. Instance: Expertise
Human: A human invocation of context provides us management, however reduces the extent of automation general. Instance: Slash instructions
Agent software program: Some context options are triggered by the agent software program itself, at deterministic closing dates. Instance: Claude Code hooks
How a lot: Maintaining the context as small as potential
One of many targets of context engineering is to steadiness the quantity of context given – not too little, not an excessive amount of. Although context home windows have technically gotten actually huge, that doesn’t imply that it’s a good suggestion to indiscriminately dump info in there. An agent’s effectiveness goes down when it will get an excessive amount of context, and an excessive amount of context is a value issue as effectively after all.
A few of this measurement administration is as much as the developer: How a lot context configuration we create, and the way a lot textual content we put in there. My suggestion could be to construct context like guidelines recordsdata up regularly, and never pump an excessive amount of stuff in there proper from the beginning. The fashions have gotten fairly highly effective, so what you may need needed to put into the context half a yr in the past won’t even be essential anymore.
Transparency about how full the context is, and what’s taking on how a lot house, is a vital characteristic within the instruments to assist us navigate this steadiness.
But it surely’s not all as much as us, some coding agent instruments are additionally higher at optimising context below the hood than others. They compact the dialog historical past periodically, or optimise the best way instruments are represented (like Claude Code’s Instrument Search Instrument).
Instance: Claude Code
Right here is an summary of Claude Code’s context configuration options as of January 2026, and the place they fall within the dimensions described above:
| What | Who | When to make use of / issues to know | Instance use circumstances | Different coding assistants | |
|---|---|---|---|---|---|
| CLAUDE.md | Steerage | Claude Code – At all times used at begin of a session | For many regularly repeated common conventions that apply to the entire venture | – “we use yarn, not npm” – “don’t overlook to activate the digital surroundings earlier than working something” – “once we refactor, we don’t care about backwards compatibility” | Principally all coding assistants have this characteristic of a foremost “guidelines file”; There are makes an attempt to standardise it as AGENTS.md |
| Guidelines | Steerage | Claude Code, when recordsdata on the configured paths have been loaded | Helps organise and modularise steerage, and subsequently restrict measurement of the at all times loaded CLAUDE.md. Guidelines may be scoped to recordsdata (e.g. *.ts for all TypeScript recordsdata), which suggests they are going to then solely be loaded when related. | – “When writing bash scripts, variables needs to be known as ${var} not $var.” paths: **/*.sh | Increasingly more coding assistants permit this path-based guidelines configuration, e.g. GH Copilot and Cursor |
| Slash instructions | Directions | Human | Frequent duties (evaluate, commit, take a look at, …) that you’ve a particular longer immediate for, and that you just need to set off your self, inside the principle context *DEPRECATED in Claude Code, superceded by Expertise* | /code-review/e2e-test/prep-commit | Frequent characteristic, e.g. GH Copilot and Cursor |
| Expertise | Steerage, directions, documentation, scripts, … | LLM (primarily based on talent description) or Human | In its easiest kind, that is for steerage or directions that you just solely need to “lazy load” when related for the duty at hand. However you possibly can put no matter further sources and scripts you need right into a talent’s folder, and reference them from the principle SKILL.md to be loaded. | – JIRA entry (talent e.g. describes how agent can use CLI to entry JIRA) – “Conventions to comply with for React elements” – “Methods to combine the XYZ API” | Cursor’s “Apply intelligently” guidelines have been at all times a bit like this, however they’re now additionally switching to Claude Code fashion Expertise |
| Subagents | Directions + Configuration of mannequin and set of obtainable instruments; Will run in its personal context window, may be parallelised | LLM or Human | – Frequent bigger duties which can be appropriate for and value working in their very own context for effectivity (to enhance outcomes with extra intentional context), or to scale back prices). – Duties for which you normally need to use a mannequin apart from your default mannequin – Duties that want particular instruments / MCP servers that you just don’t need to at all times have obtainable in your default context – Orchestratable workflows | – Create an E2E take a look at for every little thing that was simply constructed – Code evaluate carried out by a separate context and with a unique mannequin to offer you a “second opinion” with out the bags of your unique session – subagents are foundational for swarm experiments like claude-flow or Fuel City | Roo Code has had subagents for fairly some time, they name them “modes”; Cursor simply received them; GH Copilot permits agent configuration, however they’ll solely be triggered by people for now |
| MCP Servers | A program that runs in your machine (or on a server) and offers the agent entry to knowledge sources and different actions through the Mannequin Context Protocol | LLM | Use if you need to give your agent entry to an API, or to a instrument working in your machine. Consider it as a script in your machine with a lot of choices, and people choices are uncovered to the agent in a structured means. As soon as LLM decides to name this, the instrument name itself is normally a deterministic factor. There’s a pattern now to supercede some MCP server performance with abilities that describe easy methods to use scripts and CLIs. | – JIRA entry (MCP server that may execute API calls to Atlassian) – Browser navigation (e.g. Playwright MCP) – Entry to a data base in your machine | All widespread coding assistants help MCP servers at this level |
| Hooks | Scripts | Claude Code lifecycle occasions | If you need one thing to occur deterministically each single time you edit a file, execute a command, name an MCP server, and so forth. | – Customized notifications – After each file edit, verify if it’s a JS file and in that case, then run prettier on it – Claude Code observability use circumstances, like logging all executed instructions someplace | Hooks are a characteristic that’s nonetheless fairly uncommon. Cursor has simply began supporting them. |
| Plugins | A option to distribute all or any of these items | Distribute a typical set of instructions, abilities and hooks to groups in an organisation |
This looks like quite a bit – nevertheless, we’re in a “storming” part proper now and will definitely converge on an easier set of options. I count on e.g. Expertise to not solely soak up slash instructions, but additionally guidelines, which would scale back this desk by two rows.
Sharing context configurations
As I stated at first, these options are simply the muse for people to do the precise work and filling these with cheap context. It takes fairly a little bit of time to construct up an excellent setup, as a result of you need to use a configuration for some time to have the ability to say if it’s working effectively or not – there are not any unit assessments for context engineering. Due to this fact, individuals are eager to share good setups with one another.
Challenges for sharing:
- The context of the sharer and the receiver needs to be as related as potential – it really works quite a bit higher within a group than between strangers on the web
- There’s a tendency to overengineer the context with pointless, copied & pasted directions up entrance, in my expertise it’s greatest to construct this up iteratively
- Completely different expertise ranges would possibly want totally different guidelines and directions
- When you’ve got low consciousness of what’s in your context since you copied quite a bit from a stranger, you would possibly inadvertently repeat directions or contradict present ones, or blame the poor coding agent for being ineffective when it’s simply following your directions
Beware: Phantasm of management
Despite the identify, finally this isn’t actually engineering… As soon as the agent will get all these directions and steerage, execution nonetheless will depend on how effectively the LLM interprets them! Context engineering can undoubtedly make a coding agent simpler and enhance the likelihood of helpful outcomes fairly a bit. Nevertheless, generally folks discuss these options with phrases like “guarantee it does X”, or “forestall hallucinations”. However so long as LLMs are concerned, we will by no means be sure of something, we nonetheless have to assume in possibilities and select the best stage of human oversight for the job.