Opinionated, production-focused engineering standards for AI coding agents.
Tashtit is an open-source plugin marketplace for teams that want agents to work with the same discipline expected from experienced production engineers. It packages reviewable standards, repeatable workflows, and practical engineering knowledge for Claude Code, Codex, and GitHub Copilot. Cursor is an optional compatibility target.
The project is hosted as tashtit/marketplace. Its stable marketplace
identifier is tashtit.
Important
Tashtit is in its foundation phase. The repository structure, quality bar, and compatibility model are being established before the first stable plugin release. Nothing is currently advertised as production-ready.
Add the marketplace once per platform, then install plugins by name.
engineering-standards below is an example; any name from
Available plugins works.
claude plugin marketplace add tashtit/marketplace
claude plugin install engineering-standards@tashtitcopilot plugin marketplace add tashtit/marketplace
copilot plugin install engineering-standards@tashtitcodex plugin marketplace add tashtit/marketplace
codex plugin add engineering-standardsEvery plugin except evalkit is published to the Codex catalog; evalkit
supports Claude Code and GitHub Copilot CLI only.
Agent output should be safe to review, predictable to operate, and suitable for real repositories. Every stable Tashtit plugin is expected to be:
- Opinionated: it recommends a default instead of returning an unranked menu.
- Production-ready: it covers failure modes, verification, and operations.
- Enterprise-minded: it respects security, auditability, least privilege, change control, and repository policy.
- Portable: its core behavior is shared across supported agent platforms.
- Evidence-driven: normative guidance cites authoritative sources or clearly labels a Tashtit convention.
- Accountable: every plugin declares its behavior as acceptance scenarios, and a maturity claim above experimental requires a recorded passing review of each scenario on each claimed platform. Automation enforces that record, not the behavior itself.
| Platform | Priority | Status |
|---|---|---|
| Claude Code | Core | Foundation |
| OpenAI Codex | Core | Foundation |
| GitHub Copilot | Core | Foundation |
| Cursor | Optional | Research |
See compatibility for the adapter model and support policy.
Most of the initial backlog has shipped and is listed under Available plugins. The remaining backlog is organized around:
- reusable infrastructure snippets, beginning with Redis and connection lifecycle patterns;
- code style and maintainability.
The prioritized scope and acceptance criteria live in the roadmap.
| Plugin | Version | Maturity | Default behavior |
|---|---|---|---|
| Agent Parity | 0.1.0 | Experimental | Cross-agent instructions, MCP, skills, and plugin parity report, fixes on request |
| API Design Standards | 0.1.0 | Experimental | Async REST jobs and safe API deprecation |
| Architecture Diagrams | 0.1.0 | Experimental | C4 system context and container diagrams |
| Dependency Standards | 0.1.0 | Experimental | Evidence-gated dependency intake, updates, and removal |
| Engineering Standards | 0.2.0 | Experimental | Evidence-backed production change review |
| Evalkit | 0.4.2 | Experimental | Host-adaptive static skill review and worktree-isolated skill/model benchmarks (Claude Code + Copilot CLI) |
| Git Workflow | 0.1.0 | Experimental | Safe branches, commits, and pull-request handoff |
| GitHub Actions Standards | 0.4.0 | Experimental | Secure, reproducible CI and release workflows |
| Interview Kit | 0.1.0 | Experimental | Mock system design interviews graded against a cited concept library; enable after the build, not during |
| Logging Standards | 0.1.0 | Experimental | Secure structured production logging |
| Maturity | 0.3.1 | Experimental | Dockerfile, npm, repository-hygiene, and CI-workflow maturity evaluation, fixes on request |
| Repository Governance | 0.1.1 | Experimental | Audit repository governance, then optionally apply merge policy and rulesets |
| Repository Onboarding | 0.1.0 | Experimental | Read-only, evidence-backed repository assessment |
| TypeScript Library Build | 0.1.0 | Experimental | Dual ESM and CJS library build and publish |
| TypeScript Style Standards | 0.1.0 | Experimental | Convention-first TypeScript style: types, readonly, parameters, naming |
Experimental plugins are published for evaluation and do not carry Tashtit's stable compatibility or production-readiness claim.
plugins/<plugin-name>/
├── skills/ # Shared canonical behavior
├── .claude-plugin/plugin.json # Canonical Claude/Copilot manifest
└── .codex-plugin/plugin.json # Generated for Codex's required path
.claude-plugin/marketplace.json # Canonical Claude/Copilot catalog
.agents/plugins/marketplace.json # Generated Codex catalog
Tashtit reuses one file wherever platforms accept the same standard location.
When a format or a required location cannot be shared, provider files are
generated with npm run sync and checked for drift in CI; they are never
maintained as hand-copied implementations, and never as repository symlinks.
See the architecture for the full deduplication policy.
Tashtit uses maturity levels so installation never implies unsupported stability:
- Experimental: design is still changing; no compatibility guarantee, and no behavioral review is claimed.
- Candidate: automated validation passes and every acceptance scenario has a recorded passing review on each claimed platform at the published version.
- Stable: documented compatibility, security review, and release history.
- Deprecated: supported only for a documented migration window.
Only stable plugins may use the production-ready label. Every plugin is
currently experimental, so no behavioral review is being claimed yet; reviews
are recorded in tests/plugins/<name>/acceptance.json and npm run validate
rejects an unearned maturity claim.
Read CONTRIBUTING.md, the quality standard, the dependency policy, and SECURITY.md before proposing a plugin. Contributions are accepted under the Apache License 2.0.
For project direction, see GOVERNANCE.md. For usage questions, see SUPPORT.md.