autonomy-has-fences

Design before implementing: ground the change, have independent runners sketch at least two structurally distinct designs (caller usage first, then types…

Make a report, summary, status or progress page, chart, graph, diagram, flow, path or architecture view, timeline, incident or bug timeline, kanban or triage…

Adversarial multi-reviewer pass on a diff: independent reviewers on the current host's model get the same intent, rubric, and code-quality lens, and the lead…

Keep a live line to Figma while implementing a design: resolve the file key and node id for every frame, ask precisely for any missing URL, node id…

Design lifecycle steps, commands, and monitor loops to converge after crashes, retries, and duplicate execution.

Treat the cause rather than the symptom, and record a workaround as an explicit decision when one is unavoidable.

Design and create a new agent for Claude Code, Codex, or OpenCode: ask a few preference questions, write a tight persona with one job, one voice, and explicit…

Cite file:line, command output, or a captured artifact for every claim; treat should/presumably/likely as unfinished work.

Build the smallest rerunnable script, generator, validator, or skill that performs or proves nontrivial work.

Conditional code-style contract for TypeScript, JavaScript, JSX, and TSX in React frontend projects.

Capture or update the user's working preferences as a reusable mode skill.

Treat context as the scarce resource.