mvanhorn/last30days-skill · error · RuntimeError

Could not find a unique discovery output filename

Error message

Could not find a unique discovery output filename

What it means

The discovery output writer mirrors save_output(): it O_CREAT|O_EXCL-probes stem.ext, stem-date.ext, then stem-date-1..99.ext and raises RuntimeError if all 100+ candidates exist. Like [8] it means this exact discovery stem was written more than 100 times into the same directory on the same date.

Solutions

  1. Move or purge old discovery outputs for that stem in the directory.
  2. Give each run a distinct directory or stem (timestamped) instead of relying on the -N suffix chain.
  3. Reduce rerun frequency below ~100/day for the same stem.

Example fix

# before
OUT_DIR=discovery; loop writes stem.md every run

# after
OUT_DIR=discovery/$(date +%Y%m%d-%H%M); mkdir -p "$OUT_DIR"
Defensive patterns

Strategy: validation

Validate before calling

if sum(1 for _ in directory.glob(f'{stem}*.ext')) >= 100:
    directory = directory / datetime.datetime.now().strftime('%H%M%S')
    directory.mkdir(parents=True, exist_ok=True)

Try / catch

try:
    out = write_discovery_output(directory, stem, rendered, extension)
except RuntimeError:
    stem = f'{stem}-{os.getpid()}'
    out = write_discovery_output(directory, stem, rendered, extension)

Prevention

When it happens

Trigger: A discovery/monitor loop writing repeatedly to one directory: every candidate filename from the fixed list is taken, each FileExistsError is skipped, and the loop exhausts range(1, 100).

Common situations: High-frequency scheduled discovery runs; duplicated retry storms; multiple agents pointed at one shared output dir.

Related errors

AI-assisted analysis of mvanhorn/last30days-skill@c7460f6114 (2026-08-15). Data as JSON: /api/errors/e96475f34d4e490a. Report an issue: GitHub.

Appendix: source

Thrown at skills/last30days/scripts/last30days.py:1502

) -> Path:
    directory = Path(save_dir).expanduser().resolve()
    directory.mkdir(parents=True, exist_ok=True)
    extension = "json" if emit == "json" else "md"
    suffix_part = f"-{suffix}" if suffix else ""
    stem = f"{slugify(domain)}-discover-raw{suffix_part}"
    date_str = datetime.datetime.now().strftime("%Y-%m-%d")
    candidates = [directory / f"{stem}.{extension}", directory / f"{stem}-{date_str}.{extension}"]
    candidates.extend(directory / f"{stem}-{date_str}-{index}.{extension}" for index in range(1, 100))
    encoded = rendered.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 output:
            output.write(encoded)
        return candidate
    raise RuntimeError("Could not find a unique discovery output filename")


def _pre_run_prior_state(
    prior: dict[str, object] | None, run_ref: str
) -> dict[str, object] | None:
    """Reconstruct the queue state a topic had BEFORE this run identity
    recorded it.

    A row whose last_run_ref equals THIS run's run_ref was stamped by this
    very run's own earlier attempt (a finalize retry), so its surface_count
    already includes this run's surfacing: subtract it and keep the prior's
    covered state (covered_at intact) so the retry renders exactly like the
    first attempt did. Only when nothing remains after the subtraction AND
    the row was never covered is the topic genuinely first-ever (no prior).
    """
    if not prior or prior.get("last_run_ref") != run_ref:
        return prior
    previously = max(0, int(prior["surface_count"]) - 1)

View on GitHub (pinned to c7460f6114)