Repository navigation
corpus: Kotlin + Swift + Java JWT fixtures (toward #84) - #115
Merged
Merged
Conversation
Second batch of under-covered-language bench tasks per the CLAUDE.md
project invariant. Three more fixtures across two Phase-3 languages
(Kotlin + Swift) and a second Java slug pair, taking the corpus from
12 → 15 tasks.
## kotlin-jdbc-runtime-exec
Exercises:
- `kt-jdbc-string-template-sqli` — Kotlin string-template SQLi
- `kt-runtime-exec-template` — Runtime.exec with string template
Drops the JDBC safe variant — the matcher's AST query fires on
`prepareStatement` and `executeQuery` regardless of dynamic-SQL
context, which inflates FPs. Kept only the ProcessBuilder safe
decoy.
## swift-process-format
Exercises:
- `swift-eval-injection` — system(<var>) command-injection
Originally included `swift-string-format-tainted` too, but that's
noise_tier=noisy and fires too broadly even on literal format
strings; pulled out for a separate future PR. The `Process` shape
also turned out tricky (matcher requires single-line
`Process()...arguments=`), so the fixture uses `system(tool)` and
`system("/usr/bin/convert ...")` for the safe decoy.
## java-jwt-none-deserialize
Exercises:
- `java-deserialize-objectinputstream` — ObjectInputStream.readObject
- `java-jwt-none` — JWT verifier accepting alg=none
Uses fully-qualified `java.io.ObjectInputStream` to avoid the
import-line FP. Uses `Algorithm.none()` (the real JJWT API shape)
rather than a JSON literal — the matcher pattern only catches
bare `"alg":"none"` in source, not escaped-quote forms inside a
Java string literal (filed separately as a matcher-precision
improvement).
## Scoring
Each task: recall 1.0 on HIGH, precision 1.0, 0 decoy FPs.
Whole-bench:
task_count 12 → 15 (+3)
recall 0.9655 → 0.9706 (better)
precision 1.0 → 1.0 (unchanged)
decoy FP 0 0
Toward #84: 15 of 50 target tasks. Now covering Java/Rust/C/Kotlin/
Swift on the under-covered side, with Kotlin and Swift bench entries
that didn't exist before.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
13 of 15 tasks
noeljackson
added a commit
that referenced
this pull request
May 15, 2026
Second batch of under-covered-language bench tasks per the CLAUDE.md
project invariant. Three more fixtures across two Phase-3 languages
(Kotlin + Swift) and a second Java slug pair, taking the corpus from
12 → 15 tasks.
## kotlin-jdbc-runtime-exec
Exercises:
- `kt-jdbc-string-template-sqli` — Kotlin string-template SQLi
- `kt-runtime-exec-template` — Runtime.exec with string template
Drops the JDBC safe variant — the matcher's AST query fires on
`prepareStatement` and `executeQuery` regardless of dynamic-SQL
context, which inflates FPs. Kept only the ProcessBuilder safe
decoy.
## swift-process-format
Exercises:
- `swift-eval-injection` — system(<var>) command-injection
Originally included `swift-string-format-tainted` too, but that's
noise_tier=noisy and fires too broadly even on literal format
strings; pulled out for a separate future PR. The `Process` shape
also turned out tricky (matcher requires single-line
`Process()...arguments=`), so the fixture uses `system(tool)` and
`system("/usr/bin/convert ...")` for the safe decoy.
## java-jwt-none-deserialize
Exercises:
- `java-deserialize-objectinputstream` — ObjectInputStream.readObject
- `java-jwt-none` — JWT verifier accepting alg=none
Uses fully-qualified `java.io.ObjectInputStream` to avoid the
import-line FP. Uses `Algorithm.none()` (the real JJWT API shape)
rather than a JSON literal — the matcher pattern only catches
bare `"alg":"none"` in source, not escaped-quote forms inside a
Java string literal (filed separately as a matcher-precision
improvement).
## Scoring
Each task: recall 1.0 on HIGH, precision 1.0, 0 decoy FPs.
Whole-bench:
task_count 12 → 15 (+3)
recall 0.9655 → 0.9706 (better)
precision 1.0 → 1.0 (unchanged)
decoy FP 0 0
Toward #84: 15 of 50 target tasks. Now covering Java/Rust/C/Kotlin/
Swift on the under-covered side, with Kotlin and Swift bench entries
that didn't exist before.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
noeljackson
added a commit
that referenced
this pull request
May 18, 2026
Second batch of under-covered-language bench tasks per the CLAUDE.md
project invariant. Three more fixtures across two Phase-3 languages
(Kotlin + Swift) and a second Java slug pair, taking the corpus from
12 → 15 tasks.
## kotlin-jdbc-runtime-exec
Exercises:
- `kt-jdbc-string-template-sqli` — Kotlin string-template SQLi
- `kt-runtime-exec-template` — Runtime.exec with string template
Drops the JDBC safe variant — the matcher's AST query fires on
`prepareStatement` and `executeQuery` regardless of dynamic-SQL
context, which inflates FPs. Kept only the ProcessBuilder safe
decoy.
## swift-process-format
Exercises:
- `swift-eval-injection` — system(<var>) command-injection
Originally included `swift-string-format-tainted` too, but that's
noise_tier=noisy and fires too broadly even on literal format
strings; pulled out for a separate future PR. The `Process` shape
also turned out tricky (matcher requires single-line
`Process()...arguments=`), so the fixture uses `system(tool)` and
`system("/usr/bin/convert ...")` for the safe decoy.
## java-jwt-none-deserialize
Exercises:
- `java-deserialize-objectinputstream` — ObjectInputStream.readObject
- `java-jwt-none` — JWT verifier accepting alg=none
Uses fully-qualified `java.io.ObjectInputStream` to avoid the
import-line FP. Uses `Algorithm.none()` (the real JJWT API shape)
rather than a JSON literal — the matcher pattern only catches
bare `"alg":"none"` in source, not escaped-quote forms inside a
Java string literal (filed separately as a matcher-precision
improvement).
## Scoring
Each task: recall 1.0 on HIGH, precision 1.0, 0 decoy FPs.
Whole-bench:
task_count 12 → 15 (+3)
recall 0.9655 → 0.9706 (better)
precision 1.0 → 1.0 (unchanged)
decoy FP 0 0
Toward #84: 15 of 50 target tasks. Now covering Java/Rust/C/Kotlin/
Swift on the under-covered side, with Kotlin and Swift bench entries
that didn't exist before.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Toward #84.
Second batch of under-covered-language bench tasks. Bench goes
from 12 → 15 tasks. Now covering Java/Rust/C/Kotlin/Swift on the
under-covered side; Kotlin and Swift didn't have a single bench
entry before this batch.
Tasks
`kotlin-jdbc-runtime-exec` (Phase-3) — `kt-jdbc-string-template-sqli`
the matcher's AST query is too broad and FPs on it; ProcessBuilder
safe decoy retained.
`swift-process-format` (Phase-3) — `swift-eval-injection`
via `system()`. Originally bundled the noisy
`swift-string-format-tainted` matcher; pulled out because it
fires too broadly even on literal format strings.
`java-jwt-none-deserialize` (Phase-2) — `java-deserialize-objectinputstream`
The bare `"alg":"none"` pattern doesn't match through Java
string-literal escaping (filed separately as a matcher precision
improvement).
Scoring
Each task standalone: recall 1.0, precision 1.0, 0 decoy FPs.
Whole-bench:
Toward #84
15 of ~50 target tasks. Remaining under-covered: more Java/Rust
slug variants (axum-no-auth, jvm spring-cors-wildcard,
java-xxe...), PHP and C# (zero matcher pack today, may need
matcher PRs first), more Kotlin (intent/keystore/coroutine slugs),
more Swift (keychain, NSAllowsArbitraryLoads, WKWebView).
🤖 Generated with Claude Code