| name | prompt-better |
| description | Route explicit prompt compilation, lint, checkpoint, and local governance requests to the smallest PromptBetter MCP tool. |
PromptBetter
Before tool-heavy work, give a one- or two-sentence preamble naming the intended outcome. During long work, report only material phase outcomes and offer a bounded checkpoint after major milestones.
Use the smallest applicable tool:
improve_prompt: refine explicit task intent; stop at the selected execution policy.create_goal_prompt: produce a goal artifact; do not start it.create_review_fix_prompt: compile existing review findings for an exact head.lint_prompt: inspect one candidate without rewriting or executing it.get_checkpoint: retrieve latest only from this MCP session.audit_session: require explicit consent and configured local source kinds.audit_project: require explicit consent for one bounded local
portfolio/project/task/trajectory window; never apply its recommendations.render_governance_report: render the exact audit reference from this session.
Preserve commentary/final phases and user authorization boundaries. Treat compacted state as opaque. Never paste broad conversation history, start collectors, retain raw prompts, enable telemetry, call a second model/API, spawn subagents without explicit delegation, or change Codex model, reasoning effort, verbosity, fast mode, permissions, approvals, global instructions, configuration, hidden flags, or feature gates.
Unsupported host controls remain host-owned and unknown unless directly observable. Present the bounded result, warnings, provenance, and next authorized action; then stop at the requested boundary.
