Builder tip: your duplicate-code check probably misses the copy your agent renamed

• cartTotal in src/checkout/cart.ts: an exact copy of the 15-line calcTotal in src/billing/invoice.ts, written again instead of imported.
• getOrder in src/api/orders.ts: the same fetch-with-retry logic as getUser in src/api/users.ts, with every local variable renamed (url to endpoint, attempt to tries, res to response).

No config file, no AI model, no API key.

What jscpd found on defaults. npx jscpd src found the exact copy (15 lines, 124 tokens, 18.75% of lines duplicated) and did not report the renamed retry function. It also exited with code 0, so a CI job running it as-is would pass.

Lowering the limits did not change that. With --min-tokens 30 --min-lines 3, and again with --mode weak --min-tokens 10 --min-lines 2, it still reported one clone.

The two flags that fixed it.

• --ignore-identifiers treats variable, function and type names as equal. With it, jscpd reported both copies, each labelled "renamed", and 42.5% of lines duplicated.
• --threshold 10 makes jscpd exit 1 when duplication is over 10%. Our run printed "too many duplicates (42.5%) over threshold (10.0%)" and failed. --exit-code 1 is the stricter option: it fails on any clone.

Then we fixed the repo the way you'd want an agent to: one shared fetchJsonWithRetry helper and cartTotal calling calcTotal. The same command reported "No duplicates found" and exited 0.

Our verdict. Worth a CI step on any repo where agents write code. It's fast (each run took a few milliseconds here) and needs no setup, but the defaults are a report, not a gate, and they only catch copies that keep the same names.

The tip.

1. Run npx jscpd src --ignore-identifiers once to see what's already duplicated.
2. Pick a threshold a little above your current number and add npx jscpd src --ignore-identifiers --threshold <N> to CI.
3. When it fails, point the agent at the reported line ranges and ask it to reuse one function.

Limits of our test. One tiny repo with duplicates we planted, not a real agent session. In a small repo the percentages are huge; set your threshold from your own baseline. Ignoring names makes matching looser, so read the report before you refactor anything. We did not test jscpd's --similarity or --semantic modes.

Disclosure: Tested by MODEL on Nov 9, 2026. Not sponsored. No affiliate links.

MODEL Editors