| name | safe-execution |
| description | Keep code changes bound to an approved plan, avoid unrelated refactors, respect off-limits files, and summarize changed files. Use during Execute Mode. |
Safe Execution
Execute only the approved plan. Restate expected files and allowed new file locations before editing. Prefer surgical edits. Do not create arbitrary repository-root files; a root file requires an exact approved root path. Inspect project conventions and use established source, test, docs, config, script, or feature-local directories. If a current-task-created file is misplaced, preserve and move it to the correct approved path instead of deleting it. Treat untracked or unexpected files as possibly user-owned and stop/report before moving or deleting them. Do not commit, push, switch branches, install dependencies, deploy, or run destructive commands. Stop and ask if the approved plan is insufficient.
Mermaid diagrams are rendered by Workflow Suite in a uniform dark-mode visual style. For user-facing workflows, export/share paths, request lifecycles, architecture, data flow, multi-step sequences, state transitions, dependencies, validation flow, or implementation phases, include a meaningful Mermaid diagram plus concise prose unless the user requested prose only or the response is trivial. Use concise labels and the right diagram type; do not hardcode random style/classDef/light-theme overrides unless the user explicitly asks.
Professional Constraints
- Check project instructions before recommendations when the task touches code, docs, settings, or workflow behavior.
- Keep scope bounded to the user request and approved workflow phase.
- Do not print secrets, credentials, tokens, auth/session values, private runtime state, or
.env contents.
- Avoid destructive commands, deploys, pushes, resets, cleans, database mutations, and dependency installs unless explicitly approved outside this workflow.
- Prefer concise, evidence-based output with exact files or commands reviewed.
Skills vs Agents
Use this skill as in-process guidance for the current model. Use a sub-agent only when the task benefits from an isolated context window, parallel read-only research, forced preflight, or independent validation. Do not use both this skill and a same-purpose agent for the same narrow job unless the workflow or user explicitly requires it.