From ad22eee20e8e5848fa7cb874a763882fe48e721c Mon Sep 17 00:00:00 2001 From: Tibo De Peuter Date: Thu, 10 Sep 2026 16:29:13 +0200 Subject: [PATCH] feat(copilot): add personal workflow instructions --- copilot/instructions/workflow.instructions.md | 37 +++++++++++++++++++ 1 file changed, 37 insertions(+) create mode 100644 copilot/instructions/workflow.instructions.md diff --git a/copilot/instructions/workflow.instructions.md b/copilot/instructions/workflow.instructions.md new file mode 100644 index 0000000..e41f77e --- /dev/null +++ b/copilot/instructions/workflow.instructions.md @@ -0,0 +1,37 @@ +# Personal workflow instructions (Copilot CLI) + +Standing habits for how I like implementation work approached in this CLI, independent +of any one repository. Repo-specific facts and conventions belong in that repo's own +`.github/copilot-instructions.md` — this file is about *how I like to work*, not about +any particular codebase. + +## Planning + +- Use `/plan` for anything beyond a trivial change. Start the plan with branch + worktree + setup (follow the target repo's own naming/location conventions if it has them). +- Structure the plan into one or more phases, each with zero, one, or more commits, and + give each planned commit a determined Conventional Commit message up front. +- Use `/grilling` before committing to a plan whenever there's real ambiguity to resolve — + don't skip it just to move faster on non-trivial scope decisions. +- After I approve a plan, restate the exact in-scope deliverables and keep anything + discussion-only (ideas, "what if", tangents) out of scope unless I explicitly promote it. + +## Execution + +- Prefer sized, reviewable commits over one big end-of-task commit — commit at the end of + each phase, including when a fleet of sub-agents executes the phases (each phase agent + commits its own work before reporting done). +- **Model/agent choice per phase is up to the executing agent(s), not pre-assigned by the + planner.** Each phase should pick the most cost-effective model/agent capable of doing + that phase's work well, and be able to explain the choice if asked. See `model-guide.md` + in this same directory for the current model comparison — don't inline that table here, + since it changes independently of this workflow. +- Dispatch parallelizable phases with `/fleet` once the plan is approved. +- End coding work with a short state summary: changed file count, staged vs. unstaged, + branch ahead/behind, and the recommended next action. + +## PRs + +- Open a PR once the planned work is complete, unless told otherwise. +- Follow the target repo's own commit/PR conventions (Conventional Commits, merge-vs-squash, + etc.) — those live in the repo, not here.