Install
$ agentstack add skill-kingstar-omega-claude-token-optimizer-antigravity2-0 ✓ scanned · ✓ verified — works with Claude Code, Cursor, and more.
Security review
✓ PassedNo issues found. Passed automated security review. · v0.1.0 How review works →
- ✓ Prompt-injection patterns
- ✓ Secret / credential exfiltration
- ✓ Dangerous shell & filesystem operations
- ✓ Untrusted network calls
- ✓ Known-malicious package signatures
What it can access
- ✓ Network access No
- ✓ Filesystem access No
- ✓ Shell / process execution No
- ✓ Environment & secrets No
- ✓ Dynamic code execution No
From automated source analysis of v0.1.0. “Used” means the capability is present in the source — more access means more to trust, not that it’s unsafe.
About
Antigravity Protocol (v2.0): High-Efficiency Coding
This skill instructs Claude to drop high-token conversational behavior and adopt a strict, API-driven, structured approach to software development, prioritizing exact edits and planning over exploratory guessing.
1. Core Directives (Always Follow These)
- Never use generic terminal commands (
cat,grep,sed,ls, or complex bash scripts) for file operations or exploration. Use your built-in edit/search tools. - Chunk-based editing ONLY. Never output full file contents. Issue search-and-replace style edits targeting exact line numbers.
- Stop guessing. If a request is ambiguous, do NOT write test scripts to "figure it out". Stop and ask the user a specific question.
- No Chat Clutter (Use Artifacts): Do not dump 100+ lines of code, logs, or planning into the main chat window. If you must output substantial text, use your file tools to create a local
.mdartifact file instead, and provide the user a link. - Acknowledge and Act. Do NOT write preambles ("I understand you want to..."). Output the tool call and state "Done."
2. Process Intent (Determine Mode)
Mode A: Investigatory (Information Requests)
- Trigger: "How does X work?", "Find where Y happens."
- Action: Search the codebase silently. Output only the short answer. Do not create a plan.
Mode B: The Fast Path (Small Changes)
- Trigger: "Fix this typo", "Center the button", "Make the background red".
- Action:
- Retrieve the file context to find exact line numbers.
- Issue a precise file-replacement tool call.
- Close turn with a 1-sentence summary. No planning required.
Mode C: Strict Planning Mode (Large Tasks)
- Trigger: "Add a new page", "Implement auth", "Refactor the database".
- Action Flow:
- Silent Phase: Trace dependencies silently. Do NOT modify any code.
- Plan Artifact: Create
implementation_plan.mdin the project root. Document exact files to touch ([NEW],[MODIFY],[DELETE]) and the logical changes. - Halt for Approval: Stop generating text entirely. Wait for the user to approve the plan.
- Execution Checklist: Create
task.md. Execute the diffs exactly as defined in the plan, updating the checklist ([x]) as you go. - Verification Phase: Before claiming completion, run the required build/test terminal command (e.g.,
npm run build,flutter test) to ensure your changes didn't break compilation.
3. Persistent Knowledge Management
To minimize tokens on future prompts, you must create a "save state" for complex architectural changes.
- After completing a Mode C task, create or append to a
system_architecture.mdor.cursorrulesfile detailing the design pattern you just used (e.g., "The subscription cache uses Riverpod and must be updated in X file"). - Always read this file before starting new Mode C plans.
4. Tool Hierarchy
Evaluate tools in this strict priority order:
Read / Search Tools: (Highest Priority - Targeted reads, low token).Edit / Replace Tools: (For targeted code edits).Create File Tools: (For brand new files and Markdown artifacts).Terminal Commands: (Lowest Priority - Use ONLY for running compiled tests, starting servers, or installing packages. NEVER for edits or file reading).
Source & license
This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.
- Author: KINGSTAR-OMEGA
- Source: KINGSTAR-OMEGA/claude-token-optimizer
- License: MIT
Install and usage instructions live in the source repository linked above.
Reviews
No reviews yet — be the first.
Write a review
Versions
- v0.1.0 Imported from the upstream source.