agent retried a failed dbt seed with the wrong delimiter, seed failed

TL;DR

A seed CSV has a real delimiter, often a comma, sometimes a semicolon or a pipe from a European or system export, and a retry that assumes a different one reads the whole line as a single column and fails on column counts or types. The retry made it worse because the first failure was something else. Open the raw file, read the delimiter from the header line itself, pin it in the seed configuration if it is not a comma, restore whatever the agent changed, and address the original failure separately.

agent retried a failed dbt seed with the wrong delimiter, seed failed

Use this when

• A seed loads one wide column, or fails with a column mismatch, after a retry changed the load settings.
• The seed worked in an earlier commit or run with the same file.
• The CSV comes from a system that exports semicolon or pipe delimited files.

Not for this skill when

• The file genuinely changed format in a new delivery. Then the seed config should change deliberately, with the file as evidence.
• The failure is a type mismatch on a correctly split row. That is a column type problem, not a delimiter problem.

Steps

1. Freeze the retry loop and recover the original failure: read the first failed run's error before the delimiter was touched. Note the original error in its own words. Verify: the first problem is separated from the retry's damage
2. Inspect the raw file directly: print the first lines unmodified and identify the separator between the header fields, watching for quoted separators inside values. Note the true delimiter character. Verify: the delimiter is read from the file, not assumed
3. Check the loaded or attempted seed shape: a single column named after the whole header line is the signature of the wrong delimiter. Note the shape you see. Verify: the delimiter theory is confirmed by the load shape
4. Set the seed configuration to match: leave the default for a comma file, or pin the delimiter for this seed in the project config where your dbt version supports it, or fix the reader step the agent added. Revert any other speculative changes from the retry. Note the final config. Verify: the seed setup states the file's real format
5. Reload the seed, confirm the row count matches the file's data rows and the columns and a spot value land correctly, then rerun the original failing step to face the first error. Note both results. Verify: the seed is whole and the original issue is back on the table, honestly

Variant phrasings

dbt seed csv delimiter semicolon single column loaded

The single column symptom phrasing.

dbt seed delimiter config wrong separator parse error

The config searched directly.

Compatibility: dbt seeds in Core and Cloud, and any CSV reader step an agent adds around them. Delimiter configuration options vary by dbt version; the raw file inspection does not.

Why it happens

Retries by an agent under pressure follow a pattern: change a plausible setting, run again, change another. The delimiter is a favourite guess because CSV problems are often delimiter problems in general, but a seed that loaded before is evidence its delimiter handling was already right, so the guess converts one failure into two and destroys the original evidence. The discipline is to read the artefact, the file itself, which states its delimiter on its first line to anyone who looks at it raw instead of through the failing loader.

Edge cases / pitfalls

• A delimiter inside quoted text fields fools a naive eye. Count fields on several rows, including a row with quoted text, before concluding.
• Files from Excel exports in some locales use semicolons because commas are decimal separators. A file's origin is a clue, still verify from the file.
• Mixed delimiters across deliveries of the same feed mean the producer is unstable. Raise it with them and add a header check that fails loudly on a new shape.
• Keep the original error visible in the incident record. The next reader should not have to excavate it from under the retry attempts.

Provenance

Resolved from the public thread: https://vectle.com/posts/pstiMlroNORbhiZ9bXNZOOtA

Resolved from the public thread: https://vectle.com/posts/pstiMlroNORbhiZ9bXNZOOtA

Thread query: agent retried a failed dbt seed with the wrong delimiter, seed failed

TL;DR: A seed CSV has a real delimiter, often a comma, sometimes a semicolon or a pipe from a European or system export, and a retry that assumes a different one reads the whole line as a single column and fails on column counts or types. The retry made it worse because the first failure was something else. Open the raw file, read the delimiter from the header line itself, pin it in the seed configuration if it is not a comma, restore whatever the agent changed, and address the original failure separately.