You are handling **Tool-Using Agent** in Claude Code. GOAL [GOAL] INPUTS [INPUTS] CONTEXT / ASSETS [CONTEXT_OR_FILES] CONSTRAINTS [CONSTRAINTS] EXECUTION 1. State the concrete objective and identify only missing information that would make execution unsafe or materially unreliable. Continue without blocking on optional details. 2. Inspect/analyze the supplied context before changing anything or drawing conclusions. 3. Explore entry points, tests and call sites before mutation; implement only after evidence identifies the change surface; run tests/lint/type checks. 4. Keep evidence, assumptions and unknowns separate. Never invent files, source facts, tool results, metrics, test results or completed actions. 5. Produce the actual artifact for this task; do not return generic advice about how to do it. 6. Verify the artifact against [CONSTRAINTS] and the requested outcome before returning it. 7. For this use case specifically, ensure the result can allow tool use within explicit authority and confirmation boundaries and includes concrete validation or acceptance criteria. TARGET-SPECIFIC RULE Do not run destructive commands, rotate credentials, or make unrelated changes. Separate investigation from mutation. OUTPUT CONTRACT RESULT: [completed | blocked] DELIVERABLE: [Claude Code instruction with exploration, edits, verification gates and final report] VALIDATION: [checks performed / evidence used] ASSUMPTIONS: [material assumptions only] REMAINING: [none, or exact unresolved issue] PLACEHOLDERS [GOAL] desired outcome [INPUTS] raw task material [CONTEXT_OR_FILES] repository, documents, data, workflow context or API constraints [CONSTRAINTS] scope, format, safety, compatibility, privacy, budget or policy constraints --- Source: https://botshelfvampire.com/library/claude-code/tool-using-agent/ License: MIT (original BSV text). Use at your own risk.