Release v0.21.0

[emal-avala]

Summary

Cuts v0.21.0. Bumps crates/lib, crates/cli, crates/eval (path-dep only), and npm/package.json from 0.20.0 -> 0.21.0. Standardizes release docs so future release branches, PR titles, PR bodies, labels, verification, and tag steps match prior release PRs.

The remaining release-prep items are stamping CHANGELOG.md and updating the two workspace package versions in Cargo.lock from 0.20.0 -> 0.21.0.

Highlights

Theme onboarding and auto-detection (#265, #268):

• Adds first-run onboarding with a live theme picker, colorblind palettes, ANSI-only palettes, and /theme reuse.
• Adds OSC 11 terminal background detection with BT.709 luminance classification, TTY gating, DA1 sentinel handling, and a hard timeout for auto theme resolution.
• Removes runtime process-wide environment mutation from theme detection.

Plugin, output-style, and settings foundations (#255, #256, #262):

• Adds settings migrations with atomic rewrites and rotating backups.
• Loads output styles from project/user disk locations and wires /output-style to disk-loaded styles.
• Adds the plugin marketplace MVP with native and compatibility loaders, /plugin management commands, and runtime-only MCP registrations.

Model-callable operations and task surfaces (#254, #259, #263):

• Adds CronCreate/CronList/CronDelete/RemoteTrigger tools for scheduled routines.
• Adds Brief, Config, and McpAuth model-callable tools with scoped validation and allow-lists.
• Adds typed task kinds plus per-kind executor registry dispatch.

Memory and safety work (#257, #258):

• Adds team memory under .agent/team-memory/ with /team-remember and read-but-not-silently-write semantics.
• Splits Bash tool safety logic into focused modules covering destructive classification, command semantics, sandbox decisions, read-only validation, and sed -i checks.

Windows and CI stability (#261, #264, #267):

• Fixes cross-platform remote-trigger compilation, cfg-gates Unix-only tests, fills missing ToolContext fields, and makes config-path tests hermetic on Windows.

Full changelog will be in CHANGELOG.md under the [0.21.0] heading before merge.

Verification (RELEASING.md section 4)

• PR title is Release v0.21.0
• Branch is release/v0.21.0
• run-e2e label added
• cargo check --all-targets (GitHub Actions CI run 945, prior head)
• cargo clippy --all-targets -- -D warnings (GitHub Actions CI run 945, prior head)
• cargo fmt --all -- --check (GitHub Actions CI run 945, prior head)
• cargo test --all-targets (GitHub Actions CI run 945, Ubuntu + Windows, prior head)
• Release artifact builds passed for Linux, macOS, and Windows in GitHub Actions CI run 945 (prior head)
• GitHub Actions CI passed for the latest head
• E2E workflow from run-e2e label passed
• CHANGELOG.md stamped for v0.21.0
• Cargo.lock workspace package versions bumped to v0.21.0

Local cargo commands were not run because the local sandbox wrapper fails before shell startup with unexpected argument '--sandbox-policy'.

After merge

Follow RELEASING.md section 7: tag v0.21.0 on main and push. Release automation handles Linux/macOS/Windows binaries, crates.io publish, npm publish, Docker image publish, and Homebrew tap update.

[emal-avala]

Draft CHANGELOG.md section for the release stamp:

## [0.21.0] - 2026-05-04

### Added

- **Plugin marketplace MVP** (#262): disk-loaded local plugins with native and compatibility formats, `/plugin list|enable|disable|install|remove`, project enable state, and runtime-only MCP registrations.
- **First-run onboarding and themes** (#265, #268): first interactive launch now shows a welcome flow with a live theme picker, colorblind and ANSI-only palettes, `/theme`, and OSC 11 terminal background detection for `auto`.
- **Model-callable tools** (#254, #259): CronCreate/CronList/CronDelete/RemoteTrigger for scheduled routines, plus Brief, Config, and McpAuth tools with scoped permissions and allow-lists.
- **Settings migrations and output styles** (#255, #256): atomic settings migrations with rotating backups, disk-loaded output styles from project/user config dirs, and prompt-cache invalidation by active style.
- **Team memory and task kinds** (#257, #263): version-controlled team memory under `.agent/team-memory/`, `/team-remember`, and typed task records with per-kind executor dispatch.

### Changed

- **Bash tool hardening** (#258): split bash safety logic into testable modules for destructive classification, command semantics, read-only validation, sandbox decisions, and `sed -i` path checks.
- **Roadmap updates** (#260, #266): marked shipped Phase 8 items and added cloud-runtime, kanban, app-builder, and terminal-theme follow-up plans.

### Fixed

- **Windows and CI stability** (#261, #264, #267): fixed cross-platform remote trigger child-pid compilation, cfg-gated Unix-only tests, filled missing `ToolContext` fields, and made runtime tests hermetic across Windows config path behavior.
- **Theme auto-detection safety** (#268): `auto` theme resolution no longer mutates process-wide environment variables after the Tokio runtime starts.

Also update the links at the bottom:

[Unreleased]: https://github.com/avala-ai/agent-code/compare/v0.21.0...HEAD
[0.21.0]: https://github.com/avala-ai/agent-code/compare/v0.20.0...v0.21.0
[0.20.0]: https://github.com/avala-ai/agent-code/compare/v0.19.0...v0.20.0

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 20c1755fd2

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Sync Cargo.lock with the workspace version bump

This commit bumps agent-code/agent-code-lib to 0.21.0 in manifests, but the tracked lockfile still records 0.20.0 for both workspace package stanzas (Cargo.lock entries for name = "agent-code" and name = "agent-code-lib"). That leaves the release metadata inconsistent and forces a follow-up lockfile rewrite on the next lock-resolving Cargo run, which is exactly the kind of release drift this prep commit is supposed to avoid.

Useful? React with 👍 / 👎.