mvanhorn/last30days-skill · error · RuntimeError

_save_output: could not find a unique filename after 101 att

Error message

_save_output: could not find a unique filename after 101 attempts in {path}

What it means

Raised by _save_output in hosted.py when all 101 candidate filenames (the base name plus suffixes -1..-100) already exist in the save directory. The writer uses os.open with O_CREAT|O_EXCL to guarantee atomic no-clobber creation; if every candidate collides it gives up rather than overwrite. In practice this means the same slug/date/extension combination has been saved ~100 times.

Source

Thrown at skills/last30days/scripts/lib/hosted.py:232

    extension = "json" if emit == "json" else "md"
    suffix_part = f"-{suffix}" if suffix else ""
    base = path / f"{slug}-raw{suffix_part}.{extension}"
    date_str = datetime.now().strftime('%Y-%m-%d')
    candidates = [base]
    candidates.append(path / f"{slug}-raw{suffix_part}-{date_str}.{extension}")
    for i in range(1, 100):
        candidates.append(path / f"{slug}-raw{suffix_part}-{date_str}-{i}.{extension}")
    encoded = content.encode("utf-8")
    for candidate in candidates:
        try:
            fd = os.open(candidate, os.O_CREAT | os.O_EXCL | os.O_WRONLY, 0o644)
        except FileExistsError:
            continue
        with os.fdopen(fd, "wb") as f:
            f.write(encoded)
        return candidate
    # Fallback: all 101 candidates existed (extremely unlikely).
    raise RuntimeError(
        f"_save_output: could not find a unique filename after 101 attempts in {path}"
    )


def _render_complete(row: dict, topic: str, emit: str, save_dir, save_suffix: str) -> int:
    synthesis = row.get("synthesis_text") or ""
    raw_markdown = row.get("raw_markdown") or ""
    if emit == "json":
        payload = {
            key: row.get(key)
            for key in ("id", "status", "synthesis_text", "raw_markdown")
            if key in row
        }
        rendered = json.dumps(payload, indent=2, sort_keys=True)
        save_content = rendered
    else:
        # The server report is the content source; it already synthesized.
        # All markdown-ish emit modes print the synthesis text as-is.

View on GitHub (pinned to c7460f6114)

Solutions

1. Delete or archive the existing 100+ collision files for that slug in the save directory.
2. Vary the inputs: use a distinct --save-suffix, topic slug, or output directory for repeated runs.
3. Fix the caller that re-saves the identical artifact in a loop (the collision count is the symptom, the loop is the bug).

Example fix

# before: repeated identical invocations
for i in range(150):
    run(slug="my-topic", suffix="", out_dir=same_dir)  # exhausts -1..-99

# after: unique suffix per run
for i in range(150):
    run(slug="my-topic", suffix=f"v{i}", out_dir=same_dir)

from pathlib import Path

def can_save(path: Path, base_candidates: list[str]) -> bool:
    return not all((path / c).exists() for c in base_candidates)

try:
    out = _save_output(...)
except RuntimeError as e:
    if 'unique filename' in str(e):
        # switch save_dir or make the slug/suffix unique, then retry once
        out = _save_output(..., suffix=f"{suffix}-{run_token}")

When it happens

Trigger: A loop or cron job re-publishing the identical topic slug with the same save_suffix and date more than 100 times in one day; an automation bug calling _save_output in a tight loop; a hostile/accidental pre-creation of all candidate names.

Common situations: A retry storm in a hosted batch renderer; scheduled runs that regenerate the same topic hourly; someone seeded the directory with blocking filenames. The code comment itself notes it is 'extremely unlikely' in normal use.

Related errors

• save_output: could not find a unique filename after 101 atte
• [last30days] Cannot read --synthesis-file: {exc}\n
• [CompetitorsPlan] Cannot read plan file: {exc}\n
• Could not find a unique discovery output filename
• [Planner] Cannot read --plan file: {exc}\n

AI-assisted analysis of mvanhorn/last30days-skill@c7460f6114 (2026-08-15). Data as JSON: /api/errors/f9a110d82ef76ea9. Report an issue: GitHub.