Plan coding work as a compact implementation-step bundle with a conservative research decision, high-level overview first, and planning artifacts written only under .project_planning/YYYY-MM-DD_FEATURE_NAME/. This skill plans only — it never writes, edits, or refactors implementation code, and never skips ahead to building. If only the skill name is given, ask for a high-level description first before writing planning files.
Execute an approved coding plan as serial-first implementation steps with tight scope control, planner-defined verification strategy, and mandatory sub-agent/worktree execution.
Review coding changes against the approved plan, identify concrete bugs, regressions, and plan-adherence gaps, run approved isolated fix passes, then close the loop with optional planning-doc cleanup and PR/MR creation.
Execute an approved coding plan in staged steps with tight scope control, planner-defined verification strategy, deferred automated verification by default, and mandatory isolated sub-agent dispatch when the runtime supports safe worktree execution.
Close out a completed coding loop by validating final chain readiness, checking for leftover repository state, and optionally preparing and creating a PR or MR when the remote provider is supported.
Plan coding work in stages with a mandatory research-decision checkpoint, a high-level overview first, and planning artifacts written only under .project_planning/YYYY-MM-DD_FEATURE_NAME/. If only the skill name is given, ask for a high-level description first before writing any planning files.
Review coding changes against the approved plan, identify concrete bugs, regressions, and plan-adherence gaps, and, when approved, run a batched sequential review-fix loop using isolated sub-agents before finalization.
Fix grammar and clarity in any text while making the absolute minimum number of changes and preserving as much original wording as possible.