fix(review): address trail + CI findings on scope context and allowlist

86723b0·

peyton-alt·1w ago·8 files·+75 added/-15 removed

Four review findings and one lint failure, all confirmed real:

• Commit cap kept the OLDEST commits (first lines of --reverse output) and silently dropped everything near HEAD, while the prompt forbids re-deriving scope. Cap newest-first, then reverse the kept slice for reading order — recent commits are the ones a review can least afford to lose. (Cursor Bugbot, high)
• Inline diff budget 48KiB could exceed Windows' ~32KiB CreateProcess command-line cap (prompt travels as argv and ENTIRE_REVIEW_PROMPT). Lowered to 24KiB. (Copilot)
• A diff containing \` (any markdown-touching change) closed the diff fence early, letting diff content escape into instruction position. Fence is now one backtick longer than the longest run in the diff. (Copilot)
• Bash(git branch:*) allowed mutating -d/-D/-m; removed from the read-only allowlist. (Copilot)
• guidedProfileTask's profileName parameter became unused when the built-in fallback was removed; parameter dropped. (revive)

Co-Authored-By: Claude Fable 5 noreply@anthropic.com

Changes

8

• cmd/entire/cli

• agent/claudecode

  • Mreviewer.go+1/-1
  • Mreviewer_test.go+1/-1
  • review
    • Mpicker.go+2/-2
    • Mpicker_internal_test.go+4/-4
    • Mprompt.go+26/-2
    • Mprompt_test.go+22
    • Mscope.go+13/-5
    • Mscope_test.go+6

50 unmodified lines

Mcmd/entire/cli/agent/claudecode/reviewer.go+1/-1

437 unmodified lines

Mcmd/entire/cli/agent/claudecode/reviewer_test.go+1/-1

135 unmodified lines

Mcmd/entire/cli/review/picker.go+2/-2

31 unmodified lines

Mcmd/entire/cli/review/picker_internal_test.go+4/-4

130 unmodified lines

```

// skill-bearing workers kept receiving the maximal-audit brief forever.
func TestGuidedProfileTask_NoBuiltinFallbackPersisted(t *testing.T) {

t.Parallel()
if got := guidedProfileTask(DefaultProfileName, "", "", ""); got != "" {
    t.Fatalf("guidedProfileTask with nothing user-provided = %q, want empty", got)
}
}

Mcmd/entire/cli/review/prompt.go+26/-2

315 unmodified lines

```

```

Mcmd/entire/cli/review/prompt_test.go+22

```
10 unmodified lines

```

const (
    scopeMaxCommits = 50
    scopeMaxFiles = 200
    scopeMaxUncommitted = 100
    scopeDiffInlineLimit = 48 * 1024
    scopeMaxCommits = 50
    scopeMaxFiles = 200
    scopeMaxUncommitted = 100
    // scopeDiffInlineLimit stays well under Windows' ~32KiB CreateProcess
    // command-line cap (the prompt travels as one argv string and again via
    // ENTIRE_REVIEW_PROMPT), leaving headroom for the rest of the prompt.
    scopeDiffInlineLimit = 24 * 1024
)

// BuildScopeContext enumerates the review scope for prompt injection: the
// context
func buildScopeContextCapped(ctx context.Context, repoRoot, baseRef string, caps scopeContextCaps) (reviewtypes.ScopeContext, error) {
    var sc reviewtypes.ScopeContext

commitsOut, err := runGit(ctx, repoRoot, "log", "--reverse", "--format=%h %s", baseRef+"..HEAD")
    // Newest-first from git, cap (keeping the newest — the prompt declares
    // this list authoritative, and recent commits are the ones a review can
    // least afford to lose), then reverse for oldest-first reading order.
    commitsOut, err := runGit(ctx, repoRoot, "log", "--format=%h %s", baseRef+"..HEAD")
    if err != nil {
        return sc, fmt.Errorf("list scope commits: %w", err)
    }
    sc.Commits, sc.CommitsTruncated = capScopeLines(commitsOut, caps.maxCommits)
    slices.Reverse(sc.Commits)

filesOut, err := runGit(ctx, repoRoot, "diff", "--name-status", baseRef+"...HEAD")
    if err != nil {
        return sc, fmt.Errorf("list scope files: %w", err)
    }

return sc, nil
}
```

Mcmd/entire/cli/review/scope.go+13/-5

```
587 unmodified lines

```