[Extension]: Add

首先检查链接仓库中提交的 extension.yml、README.md、docs/STATUS.md 和 docs/LIMITATIONS.md。将提议的目录条目与仓库的扩展要求进行比较，然后验证安装、命令、hooks 以及记录的限制；当扩展以准确的目录元数据被接受，并且检查可重复执行时，即表示完成。

Extension ID

mainframe

Extension Name

Mainframe Modernization (COBOL to Java)

Version

1.0.1

Description

Gated, evidence-driven COBOL-to-Java modernization: deterministic discovery, specification recovery with paragraph-level traceability, domain decomposition, target architecture, data modernization with a byte-level fidelity proof, forward engineering behind a specification firewall, and golden-master equivalence verification

Author

Ankur Maan

Repository URL

https://github.com/maanankur/spec-kit-mainframe

Download URL

https://github.com/maanankur/spec-kit-mainframe/archive/refs/tags/v1.0.1.zip

License

MIT

Homepage (optional)

No response

Documentation URL (optional)

https://github.com/maanankur/spec-kit-mainframe/blob/main/README.md

Changelog URL (optional)

https://github.com/maanankur/spec-kit-mainframe/blob/main/CHANGELOG.md

Required Spec Kit Version

=0.1.0

Required Tools (optional)

python >=3.10 (required), docker >=24.0 (optional — legacy GnuCOBOL harness and the compose stack)

Number of Commands

12

Number of Hooks (optional)

4

Tags

python >=3.10 (required), docker >=24.0 (optional — legacy GnuCOBOL harness and the compose stack)

Key Features

• Deterministic discovery: inventory, dependency graph, CRUD matrix, copybook field maps from the raw COBOL/JCL/copybooks
• Specification recovery with mandatory paragraph-level traceability (100% gate) and sourced business rules
• Domain decomposition and target-architecture scoring with a visible, computed wave plan
• Data modernization with a byte-level round-trip fidelity proof (decode -> re-encode -> compare) before any schema is trusted
• Forward engineering behind a specification firewall: generated code has no read path to the legacy source
• Golden-master equivalence: runs the real COBOL under GnuCOBOL and compares field-by-field against the generated Java
• 7 hash-bound gates (G1-G7); a red evidence row can only be waived (owner, control, expiry), never approved
• Optional --auto-approve for unattended runs, still fully evidenced and audited

Testing Checklist

• Extension installs successfully via download URL
• All commands execute without errors
• Documentation is complete and accurate
• No security vulnerabilities identified
• Tested on at least one real project

Submission Requirements

• Valid extension.yml manifest included
• README.md with installation and usage instructions
• LICENSE file included
• GitHub release created with version tag
• All command files exist and are properly formatted
• Extension ID follows naming conventions (lowercase-with-hyphens)

Testing Details

Installed via specify extension add --dev locally, and via specify extension add mainframe --from <release zip URL> in a fresh Spec Kit project (non-interactive confirm). All 12 commands render into both .claude/commands and .claude/skills depending on target project layout; the modernize/assess/gate_decide/verify_source_readonly tools were run end-to-end through both the bash and PowerShell wrapper entry points. Exercised against a real legacy application: AWS CardDemo (44 COBOL programs, 870 paragraphs) - two full runs with the source Claude Code plugin (440 business rules traced at 100%, 11 bounded contexts, a Spring Boot 3 + React application built and deployed under Docker Compose, the COBOL batch chain executed under GnuCOBOL, golden-master comparison with 0 unexplained differences on sample data), then repackaged and re-verified as this extension. No secrets, credentials or telemetry in the extension; the legacy source tree is verified read-only by hash before every gate (verify_source_readonly.py). Environment: Windows 11, Git Bash, Python 3.12, specify-cli 0.16.2, Docker for the harness.

Example Usage

specify init my-project --integration claude
cd my-project
specify extension add mainframe --from https://github.com/maanankur/spec-kit-mainframe/archive/refs/tags/v1.0.1.zip

# from inside the Spec Kit project - source can live anywhere, workspace lands here
/speckit.mainframe.modernize /path/to/cobol-app
/speckit.mainframe.spec        # phase 2 -> Gate G1
/speckit.mainframe.gate --gate G1 --decision approved-with-conditions --by alice@x --condition "C1 ..."
/speckit.mainframe.status

Proposed Catalog Entry

{
  "name": "Mainframe Modernization (COBOL to Java)",
  "id": "mainframe",
  "description": "Gated, evidence-driven COBOL-to-Java modernization: deterministic discovery, specification recovery with paragraph-level traceability, domain decomposition, target architecture, data modernization with a byte-level fidelity proof, forward engineering behind a specification firewall, and golden-master equivalence verification",
  "author": "Ankur Maan",
  "version": "1.0.1",
  "download_url": "https://github.com/maanankur/spec-kit-mainframe/archive/refs/tags/v1.0.1.zip",
  "repository": "https://github.com/maanankur/spec-kit-mainframe",
  "documentation": "https://github.com/maanankur/spec-kit-mainframe/blob/main/README.md",
  "changelog": "https://github.com/maanankur/spec-kit-mainframe/blob/main/CHANGELOG.md",
  "license": "MIT",
  "category": "modernization",
  "effect": "read-write",
  "requires": {
    "speckit_version": ">=0.1.0",
    "tools": [
      {"name": "python", "required": true, "version": ">=3.10"},
      {"name": "docker", "required": false, "version": ">=24.0"}
    ]
  },
  "provides": {"commands": 12, "hooks": 4},
  "tags": ["cobol", "mainframe", "legacy-modernization", "java", "spring-boot"],
  "verified": false,
  "downloads": 0,
  "stars": 0,
  "created_at": "2026-09-11T00:00:00Z",
  "updated_at": "2026-09-11T00:00:00Z"
}

Additional Context

Packaged from a Claude Code plugin (AgenticModernizationFramework/mainframe-modernization) that was developed and exercised over two end-to-end runs against AWS CardDemo before being converted to a Spec Kit extension. docs/STATUS.md and docs/LIMITATIONS.md in the repo are candid about what is implemented versus still scaffolding (e.g. most technique skills and MCP integrations are placeholders; the deterministic pipeline, gates and golden-master verification are fully working).