diff --git a/pkg/deploy/README.org b/pkg/deploy/README.org index 9c6046f..e750a46 100644 --- a/pkg/deploy/README.org +++ b/pkg/deploy/README.org @@ -39,3 +39,49 @@ - Maintain one =sighelp_= per actual function. - Add =// gpt:= comments to express intent or relationships. - Avoid runtime logic — this file is for *structure*, not execution. + +* GPT-Assisted Spec → Code Change Workflow + + To reduce time spent mapping high-level spec changes to actual code edits, we use this workflow to integrate GPT into the loop. This allows structured delegation of search, mapping, and edit proposal. + + **Flow** + + 1. You declare a spec change as a structured Org block (see below). + 2. GPT uses =sighelp.go= (and optionally the real code) to: + - Identify affected functions + - Propose an edit plan + - Track and validate type-level constraints + 3. You confirm the plan or adjust scope. + 4. GPT writes candidate diffs or summaries for manual patching. + + **Example Change Request** + + #+BEGIN_SRC org + ,* Change: Make artifact store initialization async with retry + ,* Affects: InitMinIOArtifactStore, InitMinIOCodeQLDatabaseStore + ,* Required: non-blocking behavior, robust to transient failures + ,* Notes: Must be compatible with sighelp stubs and InitX signatures + #+END_SRC + + **GPT Responsibilities** + + - Match affected symbols from =sighelp_XXX()= stubs + - Generate patch plan as Org list: + #+BEGIN_SRC org + ,* deploy/init.go + - InitMinIOArtifactStore: wrap NewMinIOArtifactStore in goroutine, add retry + - InitMinIOCodeQLDatabaseStore: apply same pattern + #+END_SRC + - Output scoped diffs, patch instructions, or replacement code + + **Optional Enhancements** + + - GPT can update =sighelp.go= alongside implementation changes + - You may keep =change.org= files in the repo to track historical refactor plans + - Each change block can include tags like =:spec:async:init:= for search + +* Summary + + This structure treats GPT as a symbolic reasoning assistant that uses =sighelp.go= as its internal call graph. It allows high-level human changes to be mapped, tracked, and diffed without manual bottom-up spelunking. + + This flow is especially effective when multiple entry points share structural patterns (e.g. InitXXX for services).