Files
msd-core/docs/how-to/interpret-scope-conformance-warnings.md
Jakub Zych a9a7a328e6 refactor: hard-fork GSD -> MSD (Make Software Done)
Mechanical rename produced by scripts/msd-rename.cjs: gsd/Gsd/GSD -> msd/Msd/MSD
across contents and paths, upstream package/repo coordinates -> @golem15/msd-core
and golem15com/msd-core. Deep links into upstream history, sibling upstream
packages, the GSD-2 import feature, CHANGELOG.md and .changeset/ are kept as-is.

Hand edits on top: MSD block-letter banner and logos, LICENSE copyright line,
package/plugin identity, regenerated lockfile, install-tree fixtures, derived
registries and benchmark baseline; migration checksum baseline re-locked
(MSD keeps its own install state, so no install had applied the old sums);
sort-order and regex-escaped expectations in tests adjusted.
2026-10-06 01:47:40 +02:00

97 lines
6.3 KiB
Markdown

# How to interpret scope-conformance warnings
**Goal:** Understand the advisory `warnings` that the `worktree.cleanup-wave` gauntlet emits when a plan branch commits changes outside the scope it declared, and decide what — if anything — to do about one.
**Prerequisites:** MSD Core is installed and you have run `/msd-execute-phase` with worktree isolation enabled (`workflow.use_worktrees: true`, the default), so that plan branches were merged back by a `cleanup-wave`.
---
## What you will see
When `/msd-execute-phase` runs plans in isolated worktrees, each plan branch is merged back into the phase branch by the `worktree.cleanup-wave` gauntlet. If the wave manifest recorded the plan's declared scope (`files_modified`, passed as `--files` to `worktree record-agent` / `worktree create`), the gauntlet compares the branch's actual committed diff (`HEAD...<branch>`) against that declared scope and reports any committed path that falls outside it.
A realistic `cleanup-wave` result with one such warning:
```json
{
"ok": true,
"reason": "wave-cleanup-complete",
"warnings": [
{
"code": "scope_out_of_declared",
"branch": "plan-04-add-rate-limiter",
"path": "src/util/rate-limiter-cache.cts"
}
],
"entries": [
{
"branch": "plan-04-add-rate-limiter",
"status": "merged_removed",
"warnings": [
{
"code": "scope_out_of_declared",
"path": "src/util/rate-limiter-cache.cts"
}
]
}
]
}
```
`ok: true` and `status: "merged_removed"` are unchanged by the warning — the merge happened. The same warning object appears twice: once nested under the offending entry, and once aggregated into the top-level `warnings` array so you can scan for conformance issues across the whole wave without walking every entry.
---
## Why this happens
`/msd-execute-phase` gives each plan a declared scope up front (`files_modified` in the phase plan index), so that parallel plan branches can be reasoned about and merged with some confidence about what each one touched. The scope-conformance check is a post-hoc, best-effort verification of that promise: after a branch merges, the gauntlet diffs what was actually committed against what was declared, and flags any mismatch as a warning. This is issue #2596.
---
## The two warning codes
- **`scope_out_of_declared`** — emitted once per committed path that falls outside the plan's declared scope. A branch with three unexpected paths produces three of these warnings.
- **`scope_check_unavailable`** — emitted once per entry when the scope diff itself could not be computed (a `git` failure or a timeout), rather than once per path. This means conformance is **unknown** for that entry, not clean — the check deliberately distinguishes "could not verify" from "verified and clean" so an unknown result is never silently read as a pass.
---
## It does not block
Scope conformance is advisory by design. A `scope_out_of_declared` or `scope_check_unavailable` warning never changes `ok`, `reason`, the per-entry `status`, or the process exit code — the merge proceeds exactly as it would with zero warnings. There is no failure to fix here. Promoting scope conformance to a hard gate (one that blocks the merge) would be a separate, explicitly disclosed change — it is not what this check does today.
---
## What to do about one
1. Compare the reported `path` against the plan's declared `files_modified` in the phase plan index.
2. Decide which side is wrong: either the executor committed something outside its brief (over-reach), or the plan's `files_modified` under-declared what the work actually needed to touch.
3. If the plan under-declared its scope, widen `files_modified` in the plan so future waves report accurately.
4. If the executor over-reached, review that path's changes specifically — it is already merged into the phase branch, so treat the review as catching it before it travels further (into a PR, a release, or a later phase).
---
## When nothing is reported
Absence of `scope_out_of_declared` / `scope_check_unavailable` warnings is not proof that a branch stayed in scope. The check legitimately stays silent in three cases:
- **No `--files` was recorded for the plan.** Scope is unknown, so no comparison runs at all — not even a `git` call is made for that entry.
- **Every out-of-scope path is a `.planning/**/*SUMMARY.md` artifact.** These are always exempt: the executor writes them by orchestration contract, and no plan declares them as part of its scope, so flagging them would be noise rather than signal.
- **A declared pattern had no literal prefix** (for example, `*.md`). Matching is prefix-based (see below), so a pattern with no literal prefix cannot usefully bound anything — the check suppresses warnings for that entry by design, so an advisory check never cries wolf on a pattern it cannot meaningfully evaluate.
---
## Known limits
- **Glob matching is literal-prefix only.** `src/**/*.ts` matches anything under `src/`, including `src/a/b.json` — the check does not parse glob syntax past the literal prefix.
- **Renames are not detected specially, and mostly are not gated at all.** Git's rename detection runs by default, so a *pure* rename (content unchanged, similarity 100%) is reported as a single `R` entry and appears in **no** `--diff-filter=D` output. The deletions guard therefore never fires on it — before or after #3003 — and the rename lands here at the advisory, where the new path is compared against the declared scope like any other. Only a rename that also edits the file enough to fall below git's similarity threshold decomposes into a separate add and delete; that delete is a real deletion, and the guard blocks the entry unless the old path is declared in `files_deleted`. Declaring it is the fix in that case. The old path needs no `files_modified` entry either way: a declared deletion is subtracted from this check's findings, so it never shows up as out-of-scope.
- **`/msd-quick` worktrees have no plan-declared scope.** They are never checked, for the same reason as the "no `--files` recorded" case above.
---
## Related
- [CLI Tools reference — worktree commands](../CLI-TOOLS.md#worktree-commands) — the `worktree record-agent` / `worktree create` `--files` reference
- [Execute a phase](execute-a-phase.md)
- [Debug a failed execution](debug-a-failed-execution.md)
- [docs index](../README.md)