the --target refusal names neither its cause nor the form that works [✨#660] #1031

🤖 claude/anthropic claude-opus-5 (high)

Part of RSI-Software/t3code-hyprws#660.

Observed behavior

node scripts/fork-auto-rebase.ts --dry-run --target <tag> refuses with a message that names neither the cause nor the way forward:

usage: --target must be inside the clean upstream first-parent window: v0.0.41-nightly.20260915.1735

The cause is buildAutoRebasePlan (scripts/fork-auto-rebase-plan.ts:106):

const position = positions.get(sha);
if (position === undefined || position > feasibility.ffBoundary.cleanCommitCount) {
  throw new UsageError(
    `--target must be inside the clean upstream first-parent window: ${targetOverride}`,
  );
}

Two different failures share that one sentence: a tag that is not on the upstream first-parent walk at all (position === undefined), and a tag that is on the walk but past the clean fast-forward boundary (position > cleanCommitCount). The message distinguishes neither, and prints neither the boundary it compared against nor the tag's position relative to it.

The second failure also has a trivial workaround the message does not mention: omit --target entirely. The walk then plans to the newest tag on its own, and in the case above that newest tag is v0.0.41-nightly.20260915.1735, the very tag --target had just refused. A caller who wants a census against the newest tag is refused for naming it and served for saying nothing.

Minimal reproduction

1. node scripts/fork-auto-rebase.ts --dry-run --target <newest nightly tag>: the refusal above, exit 2.
2. node scripts/fork-auto-rebase.ts --dry-run --issue-json .dump/tmp/census.json: succeeds, and its census targetTag is that same tag.

Expected behavior

The refusal separates the two causes and prints the numbers it decided on:

• for a tag off the first-parent walk, say that, because no boundary is involved;
• for a tag past the boundary, print the tag's position, the clean commit count, and the commit that ends the clean window;
• where the caller's tag is the newest tag anyway, say that omitting --target reaches it.

Evidence

Refusal captured on 2026-09-15 at hyprws b22495c259 with --target v0.0.41-nightly.20260915.1735; the same run without --target produced a census whose targetTag is that tag. Source: scripts/fork-auto-rebase-plan.ts:106.

Close conditions

• The two causes behind --target must be inside the clean upstream first-parent window produce distinguishable messages.
• A boundary refusal prints the tag's position, the clean commit count, and the commit that ends the clean window.
• A refusal for a tag the walk would reach on its own says that omitting --target reaches it.

Origin: measuring RSI-Software/t3code-hyprws#1024's before/after stop census; the first attempt pinned the target tag explicitly and was refused, costing a round trip before the working form was found by reading buildAutoRebasePlan.