Build1 publisherNot yet confirmed elsewhere3 min readPublished
Two of Claude Code's five October changes are worth switching on this week
Claude Code shipped five developer-facing changes from 1 to 6 October 2026, among them Claude Mods and an effort parameter on the Agent tool. The effort setting can go on at once, and the young You should know side agent deserves a one-repository trial first.
The Engineer · Build desk

What happened
- Some built-in features now ship as mods, and the changelog's example is /diff, which a developer can disable in /plugin or replace with a custom version.
- The You should know side agent turns on with /plugin enable cc-plugin-you-should-know@builtin and applies to first-party sessions with telemetry on.
- A 2 October update, version 2.1.288, made the side agent's notes say "we", "the main agent" or "you" depending on who is responsible for the flagged item.
- The 6 October release, 2.1.292, also added a --marketplace flag on plugin install and a retry-delay environment variable for 529 overload errors.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision Adopting the effort setting needs no plugin and no trust review, so the only judgement left is which sub-agent jobs are mechanical enough to run at lower effort.
- exposure Once a team distributes plugins through --marketplace, every installed mod is code sitting in each member's prompt path, so version pinning and source review have to become team policy.
- cost The side agent's token consumption is not in the changelog, so a one-week trial has to track spend as well as how often its flags were useful.
- constraint Two developers' sessions can now behave differently because of their installed plugins, so debugging odd behaviour has to start at /plugin before the model or the prompt.
The dev.to post's author argues that the effort parameter costs nothing and cuts the bill on mechanical sub-agent work [18]. That claim transfers to another setup only under one condition. The jobs handed to a lower-effort sub-agent have to be ones where the lower setting still returns the right answer. I'd start with the sub-agent jobs whose output a test or a diff already checks, and leave the reasoning-heavy ones alone until the results say otherwise.
You should know runs alongside the main session [9]. Its job is to look for what the task is about to break [9]. The 2 October wording change is the best piece of design in the batch [12]. In the author's reading, a note addressed to "you" asks the developer to act, while one addressed to "the main agent" says Claude should fix it and the developer can leave it alone [16]. The feature is young. Its first two fixes, both inside four days of release, taught it pronouns and then to stop writing English regardless of the configured language setting [17][13].
That telemetry condition has a practical edge. On a third-party cloud provider, or with telemetry switched off, the mod may not be available at all, according to the post [11].
The trial the post proposes uses one repository with real tests and a trusted CLAUDE.md, five working days of normal work, and a text file with a line for each flag that was useful, ignorable or wrong [20]. The decision rule is plain: keep it if useful flags outnumber wrong ones, and switch it off in /plugin if it nags about things existing hooks already catch [20].
Mods can wait for most solo developers, though the design is careful. Until 2.1.287, plugins worked from the outside of Claude Code, adding commands, skills, hooks and agents [5]. That release's changelog entry, as quoted in the post, reads: "Added Claude Mods: plugins may now modify deeper behavior." [4] Two mod APIs arrived in 2.1.292 [7]. The prompt.autocomplete API lets a mod shape what the prompt box suggests. The $.tool.register API accepts an isDeferred option, letting a mod register a tool whose schema stays out of each request until Claude has a use for it [7].
The cost is trust. A plugin that only adds a slash command can do limited damage, while a mod sits in the path of everything typed and every tool call Claude makes, the author wrote, recommending readable sources, pinned versions and a short plugin list [8]. The same rule covers the new --marketplace flag [8]. The 529 retry-delay variable is narrower still: the author would set it only after actually seeing overload errors [19].
What to watch
- Whether the Claude Code changelog starts stating how many tokens the You should know side agent consumes per session.
- Further fixes to You should know in releases after 2.1.294, which would show whether the feature is settling or still changing week to week.
- Third-party mods distributed through --marketplace installs, the point at which pinned versions and readable sources move from personal habit to team requirement.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence55
- Adoption
- Insufficient
- Hype gap+10
- Incentives
- Insufficient
- Confidence50
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
Claude Code shipped five developer-facing changes between 1 and 6 October 2026.
- [2]
Version 2.1.287 on 1 October added Claude Mods and a built-in mod called You should know.
- [3]
Version 2.1.292 on 6 October added an effort parameter to the Agent tool, a --marketplace flag on plugin install and a new retry-delay environment variable for 529 overload errors.
- [4]
The changelog entry for 2.1.287 reads: "Added Claude Mods: plugins may now modify deeper behavior."
- [5]
Until 2.1.287, a plugin could add commands, skills, hooks and agents around Claude Code.
- [6]
Some built-in features now arrive as mods; the changelog uses /diff as the example, which can be disabled in /plugin or replaced with your own.
- [7]
Two mod APIs landed in 2.1.292: prompt.autocomplete lets a mod shape what the prompt box suggests, and $.tool.register registers a tool and accepts an isDeferred option, so a mod can add a tool without paying its schema cost in every request until Claude needs it.
- [8]
The author says Mods raise the trust bar: a plugin that only adds a slash command can do limited damage, while one that modifies core behaviour sits in the path of everything you type and every tool call Claude makes; install mods from sources you can read, pin versions, keep the plugin list short, and the same rule applies to --marketplace.
- [9]
The 2.1.287 changelog describes You should know as a side agent that watches your back and flags things you or Claude might miss; it runs alongside the main session and looks for what the task is about to break.
- [10]
The enable command is /plugin enable cc-plugin-you-should-know@builtin, and the changelog notes it applies to first-party sessions with telemetry on.
- [11]
If you run Claude Code through a third-party cloud provider or have telemetry switched off, the mod may not be available.
- [12]
Version 2.1.288 on 2 October changed the wording of the side agent's notes so they say "we", "the main agent" or "you" depending on who is responsible for the thing flagged.
- [13]
Version 2.1.290 on 5 October fixed a bug where the side agent wrote English regardless of the configured language setting.
- [14]
The changelog does not state how many tokens the side agent consumes.
- [15]
Claude Code's behaviour is now sometimes the result of installed plugins, which matters when a teammate's session acts differently; when debugging odd behaviour, open /plugin first.
- [16]
In the author's reading, a flag that says "you" asks you to act, while a flag that says "the main agent" tells you Claude should fix it and you can leave it alone.
- [17]
You should know received two fixes within four days of its 1 October release.
- [18]
The author recommends turning on the Agent-tool effort habit immediately because it costs nothing and cuts the bill on mechanical sub-agent work.
ReportedInsufficientSource: dev.to post author's recommendation2 sources— create a free account to open themView cited source - [19]
The author recommends setting the 529 backoff variable only if you actually see overload errors, and ignoring --marketplace until you distribute plugins to a team.
ReportedInsufficientSource: dev.to post author's recommendation2 sources— create a free account to open themView cited source - [20]
The author's trial: pick one repository with real tests and a CLAUDE.md you trust, enable the mod, work normally for five working days, log each flag as useful, ignorable or wrong; keep it if useful flags outnumber wrong ones, and switch it off with /plugin if it nags about things your hooks already catch.
ReportedInsufficientSource: dev.to post author2 sources— create a free account to open themView cited source
Sources
1 independent publisher whose own reporting we read for this story.
- dev.toClaude Code Mods and the Side Agent: What a Solo Dev Enables
1 article · October 8, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI Coding AssistantsFollow
- Developer tool pluginsFollow
Entities
- Claude CodeFollow
- Claude ModsFollow
- You should knowFollow