Chara's Toolkit

Better engineering, any agent.

22 engineering skills for coding agents, with optional workflow tooling for Codex.

Guida in italiano

Chara's Toolkit gives coding agents focused instructions for understanding repositories, finding bugs, changing code safely, and proving that a change works. Each skill is a self-contained SKILL.md folder that can be installed in a compatible Agent Skills client. The full Codex integration adds routing, six optional Mission Control roles, and release-based updates.

Choose your setup

Full toolkit

Requires Node.js 18+ and a configured Codex installation for the complete installer and routing layer.

npx --yes github:cmdr-chara/chara-toolkit setup

This installs:

• all 22 skills;
• six optional Mission Control roles;
• workflow routing and the managed AGENTS.md instructions;
• a user-level updater that follows published GitHub releases.

Prefer a global command?

npm install --global github:cmdr-chara/chara-toolkit
chara setup

codex-toolkit remains a compatibility alias for chara.

One skill only

Browse the catalog:

npx skills add https://github.com/cmdr-chara/chara-toolkit --list

Install one skill for a compatible client at user level. Replace codex with the client adapter you use:

npx skills add https://github.com/cmdr-chara/chara-toolkit \
  --skill repository-intelligence -g -a codex

This uses the portable SKILL.md format. It does not install the full toolkit routing layer, Mission Control roles, or the updater. See the portability guide for client-specific details.

Start using it

After a full setup, inspect the installation:

npx --yes github:cmdr-chara/chara-toolkit mission-control check

Then start a new Codex task and describe the result you want. The router chooses the smallest relevant specialist; you usually do not need to name a skill.

Find important bugs in this repository, fix them, and verify the changes.

This API is slow. Measure the bottleneck, improve it, and compare the results.

Review this authentication change and report only security issues that can be verified from the code.

Your project's AGENTS.md remains higher priority than general toolkit guidance. Skills guide the agent; your host still controls files, tools, network access, and approvals.

Skill catalog

Skill	Use it for
repository-intelligence	Map architecture, ownership, dependencies, hotspots, and change impact.
bug-finder	Hunt for previously unknown correctness defects and prove or retire candidates.
debugging-investigator	Reproduce a concrete failure, trace its cause, and design a focused fix.
review-and-refactor-code	Review PRs with verified findings and bounded re-review, or plan behavior-preserving refactors.
security-review	Trace trust boundaries and assess exploitable security flaws.
optimize-codebase-performance	Measure a named bottleneck and compare a bounded optimization.
codebase-evolution-controller	Plan dependency, API, schema, and framework transitions with rollback paths.
codebase-improvement-planner	Discover and rank evidence-backed repository improvements.
documentation-synchronizer	Keep user, developer, API, migration, and operations docs aligned with code.
verification-and-release	Decide what must be verified and whether an integrated change is ready to ship.
unlazy	Track completion gates and prevent missing deliverables on substantial work.
toolchain-preflight	Resolve local shell, runtime, package-manager, and browser blockers.
typescript-quality-enforcer	Improve TypeScript and JavaScript type safety and lint discipline.
production-web-builder	Build or audit production web interfaces.
screenshot-to-interface	Reconstruct maintainable interfaces from visual references.
product-design-director	Define UX, visual direction, accessibility, and responsive behavior.
mobile-architecture-director	Choose a mobile architecture from evidence and constraints.
flutter-production-builder	Build or audit production Flutter applications.
expo-react-native-builder	Build or audit production Expo and React Native applications.
delegate-with-mission-cards	Split independent work into bounded missions with explicit ownership.
multi-agent-work-coordinator	Coordinate safe parallel work across non-overlapping resources.
content-provenance-hygiene	Inspect and sanitize authorized metadata and provenance surfaces.

Evidence-first pull request review

Use review-and-refactor-code with a specific PR or diff. It now adjusts review depth to risk, validates suspected bugs against real code paths, distinguishes introduced from pre-existing issues, and stops agent fix/review cycles after three substantive rounds. Findings begin with a count, including zero when supported by the inspected scope.

The PR review protocol defines the workflow. The opt-in policy guide includes a REVIEW.md example and adoption guidance. Chara's Toolkit itself now includes REVIEW.md and a pull-request template.

This is a skill upgrade, not an automatically installed review bot, enforced GitHub protection, or human approval. The existing security and release specialists retain their responsibilities.

The machine-readable catalog is in skills/llms.txt. The six optional roles are architect-writer, builder-writer, investigator-reader, patcher-writer, pathfinder-reader, and sentinel-reader.

What works where

Capability	Full toolkit	Individual skills
Individual SKILL.md folders	Included	Installable in compatible clients
Natural-language routing across the toolkit	Codex integration	Client selects installed skills
Mission Control roles and workflow orchestration	Codex integration	Not included
Toolkit updater and managed AGENTS.md block	Codex integration	Not included

The skill folders provide instructions, references, and scripts. They do not grant runtime permissions, API keys, or network access.

Options and updates

Preview a setup without changing files:

npx --yes github:cmdr-chara/chara-toolkit setup --dry-run

Install without registering a scheduled updater:

npx --yes github:cmdr-chara/chara-toolkit setup --no-auto-update

With a global install, manage the updater with:

chara auto-update status
chara auto-update remove
chara auto-update install

Use --codex-home /path/to/codex when Codex uses a non-default home. The setup preserves instructions outside its managed AGENTS.md block and backs up managed files before replacement.

Documentation

• Guida in italiano: installazione e uso
• Portability guide · Skill catalog
• Workflow routing · Responsibility matrix
• Automatic updates · Enterprise deployment
• Application verification · Destructive-operation safety
• Evaluations · Contributing · Changelog
• Security reporting · Third-party notices

License

MIT © 2026 cmdr-chara. Third-party sources, modifications, and preserved license notices are listed in THIRD_PARTY_NOTICES.md.