Agent-config&Claude Code mods news update

Max Levitskiy released github.com's report detailing the Agent Config Standard (ACS v1) alongside a zero-dependency TypeScript loader packaged as a skill and CLI tool set. The release introduces standardized configuration patterns designed to streamline setting resolution and credential management across multi-agent environments. Tools like orchestrate and herdr vendor the v1 library directly to standardize their configuration handling, while fellow accesses settings using the v2 agent-config CLI.

This article covers the architecture of the Agent Config Standard v1, the operational features of the ml-agent-config plugin, and how development teams can implement it within Claude Code and Codex environments. Developers building subagents or managing multi-layered project credentials will gain an understanding of secret-free configuration files, interactive onboarding workflows, and cross-plugin library vendoring protocols.

What shipped in Agent-config&Claude Code mods news update

Max Levitskiy delivered the Agent Config Standard (ACS v1) specification alongside the ml-agent-config skill plugin and TypeScript runtime loader. The system establishes a standardized configuration schema that allows developer skills and subagents to manage application settings without embedding plain-text credentials into configuration files. The release includes vendor integration helpers lib/config.ts and lib/credentials.ts, enabling direct incorporation into custom agent projects. Early adopters orchestrate and herdr have vendored the core v1 library directly into their codebases, while fellow reads config data via the v2 agent-config CLI.

The installation package installs via the command /plugin install ml-agent-config@max-skills, providing immediate access to the interactive setup flow /ml-agent-config:setup. When invoked, Claude guides the user through evaluating settings layers, determining credential storage locations, writing target files, and executing a live call to verify configuration validity. Additionally, the release provides automated migration mechanisms via agent-config adopt to migrate legacy configuration files from .agents/skill-config/<name>/ paths over to the standardized .agents/config/ folder structure.

What improved in Agent-config&Claude Code mods news update

The update separates settings into three deep-merged layers while enforcing complete isolation of sensitive credentials from committed version control. Global defaults persist across all user projects, shared repository settings remain committed for team reuse, and local checkouts handle uncommitted personal overrides. Credential fields contain resolution references rather than raw secrets, resolving keys through 1Password, environment variables, gitignored dotenv files, macOS Keychain, or custom shell commands. To reduce authentication prompts, references support a cacheVar flag that binds secrets to session memory without writing raw values to disk.

On Claude Code, the dedicated claude-code/register.ts hooks module intercepts skill execution to automate initialization and authorization. Upon component expansion, the hook executes agent-config start automatically to evaluate component readiness. For 1Password references, the module completes a single authentication prompt per session and holds resolved values inside the AGENT_CONFIG_SECRETS environment variable. Codex continues using its .codex-plugin/plugin.json manifest without hook modifications, maintaining full CLI compatibility while avoiding unexpected execution changes.

What you gain from Agent-config&Claude Code mods news update

Developers gain an interactive visual management pane accessed via /agent-config, providing fine-grained control over component statuses, inherited settings, and credential mappings. The interface allows users to view effective values, switch between configuration tiers, and edit keys directly without manually editing underlying JSON files. Teams can block inherited settings by setting explicit null entries or override parent values across specific checkout scopes. The settings pane also incorporates an interactive 1Password selector that searches account vaults via op:// references, caching item names locally across sessions to maintain human-readable references without repeatedly unlocking vaults.

Plugin authors gain a stable vendoring strategy that avoids inter-plugin runtime dependencies. By embedding lib/config.ts directly into dependent modules, plugins remain operational even if secondary tools are uninstalled. To prevent version drift across codebase copies, the release includes vendor.sh sync to update internal copies and vendor.sh check to flag unauthorized local modifications during build pipelines. This architecture eliminates unexpected runtime crashes caused by missing external plugin dependencies.

How to get Agent-config&Claude Code mods news update

To install the package and update your environment configuration, run the vendor installation command inside your target terminal session:

/plugin install ml-agent-config@max-skills

To configure or switch model execution modes within the interactive shell session, invoke the model selection slash command:

/model claude-fable-5-1

To launch a pinned instance of your CLI environment tied to a specific model version, pass the model parameter during process initialization:

claude --model fable

To lock your preferred execution model as the system default across future sessions, update your global user settings file located at ~/.claagents/config/agent-config/config.json with the target configuration key.

What to watch after Agent-config&Claude Code mods news update

Development teams should monitor how third-party agent harnesses adopt the ACS v1 specification as vendor integration expands beyond orchestrate, herdr, and fellow. The migration path from legacy .agents/skill-config/ paths will require tracking as projects execute agent-config adopt commands across active development branches. Furthermore, teams using 1Password references should verify that local vault item caches in ~/.agents/config/ remain synchronized when account credentials or vault names are updated upstream.

Security administrators should audit memory exposure risks associated with the AGENT_CONFIG_SECRETS environment variable within shared build environments. Because subprocesses spawned by the primary Claude Code session inherit environment variables, custom script steps executed during a session can access cached credentials. Teams can manage this exposure by triggering /agent-config unlock to re-authenticate or invoking /agent-config forget to purge session secrets from process memory whenever elevated credential protection is required.