fix(runner): don't drop iteration 0 in the auto-finalize fallback

[Osamaali313]

What

The auto-finalize fallback in CodexRunner._reader_iter (scripts/figcopy_runner/codex.py) and ClaudeRunner._reader_iter (scripts/figcopy_runner/claude.py) silently drops iteration 0:

disk_iters = sorted(
    _parse_iter_n_from_path(p.name) or -1
    for p in workdir.glob("img_iter*.png")
)
disk_iters = [n for n in disk_iters if n >= 0]

_parse_iter_n_from_path returns an int or None. The or -1 is meant to turn the None sentinel into -1 so the next line can drop it — but 0 is falsy, so img_iter0.png parses to 0, 0 or -1 evaluates to -1, and the n >= 0 filter discards it.

Why it matters

Iterations are 0-based (the loop policy says Iterate N = 0..{max_iters-1}), so img_iter0.png is the ordinary first artifact. This block exists to salvage a run when the CLI exits non-zero mid-loop (e.g. turn budget exhausted). With the bug:

• a run whose only artifact is iteration 0 -> disk_iters == [] -> auto-finalize is skipped -> the run is reported failed and no figure.png is written;
• when iteration 0 is the floor-passing iteration, _pick_finalize_iter never sees it and a different (possibly worse) iteration ships.

The other three callers of the same helper (codex.py:1113, codex.py:1455, claude.py:794) all correctly use is None — only these two finalize sites use or -1, so it's a self-inconsistency.

only img_iter0.png present:  buggy disk_iters=[]      fixed disk_iters=[0]
img_iter0 + img_iter2:       buggy disk_iters=[2]      fixed disk_iters=[0, 2]

Fix

Mirror the sibling sites — keep only successfully-parsed indices, treating None (not 0) as the skip sentinel:

disk_iters = sorted(
    n
    for p in workdir.glob("img_iter*.png")
    if (n := _parse_iter_n_from_path(p.name)) is not None
)

Applied to both runners. No behavior change for runs whose lowest iteration is >= 1.

The scan lives inline in the _reader_iter finalize/exit block (not a standalone function), so it isn't directly unit-testable without driving the full runner — happy to extract it into a small helper with a regression test if you'd prefer.

[Copilot]

🟢 Ready to approve

The change is a targeted, correct fix for a confirmed falsy-0 bug and aligns behavior with existing sibling call sites without introducing API or control-flow risk.

This review doesn't count toward merge requirements. Sign up for the private preview to control whether Copilot approvals count.

Pull request overview

Fixes a correctness bug in the runners’ auto-finalize fallback logic where img_iter0.png could be accidentally excluded from consideration due to 0 being falsy in Python. This ensures iteration indices are treated as 0-based consistently when salvaging a run after an early non-zero CLI exit.

Changes:

• Update CodexRunner auto-finalize disk scan to keep parsed iteration indices (including 0) and skip only None.
• Update ClaudeRunner auto-finalize disk scan with the same is not None parsing guard for consistency.
• Remove the or -1 / >= 0 sentinel pattern that incorrectly dropped iteration 0.

File summaries

File	Description
scripts/figcopy_runner/codex.py	Fixes the finalize fallback’s on-disk iteration scan so img_iter0.png is not dropped.
scripts/figcopy_runner/claude.py	Applies the same iteration-0-safe parsing pattern to Claude’s finalize fallback.

Review details

• Files reviewed: 2/2 changed files
• Comments generated: 0
• Review effort level: Lite

We're testing this review assessment. Please use 👍 or 👎 to tell us if it's correct.