Opspace

Less context. Cleaner code. Safer changes.

Opspace is a production-grade skill library for AI coding agents. It gives Codex, Claude Code, GitHub Copilot, Cursor, Gemini, and other assistants the operating discipline they need to work like reliable engineering teammates.

Give agents operating discipline, not more instructions.

Why Opspace

AI coding agents are powerful, but they often waste context, read too broadly, edit outside the task, skip verification, or produce handoffs that are hard to trust.

Opspace packages practical engineering workflows as reusable skills:

• Read less, but understand enough.
• Change less, but change the right thing.
• Prefer existing project patterns.
• Keep patches focused and reviewable.
• Verify work with the smallest useful test surface.
• Explain outcomes clearly without bloating the context window.

Before / After

All comparisons use controlled paired runs on the same fixture with a rubric score. Results are measured-once unless noted.

context-budgeting — small bugfix

| | Baseline | Opspace | | --- | ---: | ---: | | Full files read | 12 | 0 | | File slices read | 0 | 5 targeted | | Files changed | 1 | 1 | | Verification passed | ✅ | ✅ |

repo-discovery — new repo orientation

| | Baseline | Opspace | | --- | ---: | ---: | | Unrelated files read | 2 | 0 | | Command confirmed from repo | ❌ guessed | ✅ package.json | | Checkout files found | 1 (impl only) | 2 (impl + tests) | | Assumptions left | 2 | 0 |

clean-code-guardian — scope control

| | Baseline | Opspace | | --- | ---: | ---: | | Unrelated edits (rename, reformat) | 2 | 0 | | Speculative abstractions added | 1 | 0 | | Patch reviewable as one change | ❌ | ✅ | | Verification | ✅ | ✅ |

safe-refactor — extract total calculation

| | Baseline | Opspace | | --- | ---: | ---: | | Behavior lock written before edit | ❌ | ✅ | | Public contract accidentally changed | ✅ broke | ❌ preserved | | Tests pass after refactor | ❌ | ✅ |

test-strategy — focused verification

| | Baseline | Opspace | | --- | ---: | ---: | | Command chosen | pnpm test (all) | pnpm test:checkout (focused) | | Smallest useful check selected | ❌ | ✅ | | Remaining risk stated | ❌ | ✅ |

git-discipline — atomic commits

| | Baseline | Opspace | | --- | ---: | ---: | | Commits created | 1 (mixed) | 3 (atomic) | | Conventional message prefix | ❌ | ✅ | | Checked diff before staging | ❌ | ✅ | | Commit Ledger filled | ❌ | ✅ |

Full traces: benchmarks/results/

Hero Skills

Opspace starts small on purpose. The first skills focus on agent behavior that matters in almost every software project.

| Skill | Purpose | | --- | --- | | context-budgeting | Reduce wasted context by reading progressively and caching useful repo facts. | | repo-discovery | Map project structure, commands, dependencies, and ownership before changing code. | | clean-code-guardian | Keep patches idiomatic, scoped, readable, and aligned with local patterns. | | safe-refactor | Refactor in small behavior-preserving steps with verification gates. | | test-strategy | Choose the right test level and run the smallest meaningful checks. | | git-discipline | Maintain atomic commits, write Conventional Commit messages, and prevent secret leaks. | | pr-handoff | Write clear PR descriptions — what changed, why, how to test, what risk remains. |

Seven hero skills are now available. The first 6 have benchmark cases and manual eval harnesses.

More skills will be added only when they earn their place.

Agent Support

| Agent | Status | Install | | --- | --- | --- | | Claude Code | ✅ Plugin | /plugin install hammond01/opspace | | Antigravity | ✅ Extension | agy plugin install https://github.com/hammond01/opspace | | Cursor | ✅ Adapter | copy adapters/cursor/opspace.mdc → .cursor/rules/ | | GitHub Copilot | ✅ Adapter | copy adapters/copilot/copilot-instructions.md → .github/ | | Gemini CLI | ✅ Adapter | copy GEMINI.md to project root |

Install

One line. Find every agent. Install for each.

# macOS / Linux / WSL / Git Bash
curl -fsSL https://raw.githubusercontent.com/hammond01/opspace/main/install.sh | bash

# Windows (PowerShell 5.1+)
irm https://raw.githubusercontent.com/hammond01/opspace/main/install.ps1 | iex

~30 seconds. Needs Node >=18. Skips agents you don't have. Safe to re-run.

Or with npx directly:

npx opspace add

Manual

Copy the adapter for your tool:

| Tool | Source | Destination | | --- | --- | --- | | Claude Code | /plugin install hammond01/opspace | — | | Cursor | adapters/cursor/opspace.mdc | .cursor/rules/opspace.mdc | | GitHub Copilot | adapters/copilot/copilot-instructions.md | .github/copilot-instructions.md | | Gemini CLI | GEMINI.md | project root GEMINI.md | | Codex | AGENTS.md | project root AGENTS.md |

Project Shape

opspace/
├── docs/          # install guides, agent compatibility, skill format, roadmap
├── skills/        # canonical Opspace skills
├── adapters/      # agent-specific usage instructions
├── benchmarks/    # task cases, methodology, results
├── evals/         # manual eval prompts and rubrics
└── fixtures/      # small repos used by benchmark cases

Benchmarks

Opspace earns trust with receipts, not vibes. All results use controlled paired runs with rubric scoring.

| Benchmark | Status | Result | | --- | --- | --- | | context-budgeting | measured-once | view | | repo-discovery | measured-once | view | | clean-code-guardian | measured-once | view | | safe-refactor | measured-once | view | | test-strategy | measured-once | view | | git-discipline | measured-once | view |

• Measurement Protocol
• Result Format
• Eval Harnesses

Roadmap

v0.1

• README and docs foundation.
• First 6 hero skills (context-budgeting, clean-code-guardian, repo-discovery, safe-refactor, test-strategy, git-discipline).
• Manual eval harnesses for all 6 hero skills.
• Codex and Claude Code support.
• Benchmark methodology.

v0.2

• Cursor, Copilot, and Gemini adapters.
• Sample benchmark results.
• More production-readiness skills.

v0.3

• Installer prototype.
• Dry-run and uninstall support.
• Skill validation helpers.

Contributing

Opspace is designed to be small, practical, and strict about signal quality.

Good contributions:

• Improve a skill with clearer behavior.
• Add a realistic benchmark case.
• Add support for an agent with precise install instructions.
• Remove unnecessary wording.
• Make claims more measurable.

Avoid:

• Generic advice that an agent already knows.
• Large vague checklists.
• Vendor-specific lock-in.
• Skills that exist only because the name sounds useful.

See CONTRIBUTING.md for contribution guidelines.

License

MIT. See LICENSE.