NewYour coding agent can read the release notes before it upgrades.Set up the MCP server →
PyPI
PyPI, #771 most downloaded on PyPI
Lightweight static analysis for many languages. Find bug variants with patterns that look like source code.
Latest stable
1.185.0
Last release 12 days ago
10 Nov 2026
Ships on a steady schedule
a new release about every 2 weeks
Nearly every release is documented
notes for 54 of the last 60 stable releases
Nothing withdrawn
no release was ever pulled
7 years old
368 releases · first in 2020
Drag across the ruler to compare, or tap a release to open it.
PHP: Improved virtual call resolution for basic trait composition. (CODE-9176)
requirements.txt dependency declared as a version range (litellm>=1.10) or with no version (flask) now produces a finding for every advisory whose affected versions overlap the declared range, in both the Python and native resolvers. Such a finding has no single installed version to confirm, so it needs review. A range line next to a pinned copy of the same package produces no finding of its own, since the pinned copy names the installed version. (SC-4243)SEMGREP_LOG_JSON_FILE to emit structured execution events with scan metadata alongside existing text output. (SMS-3670)Scans are slightly faster. The dataflow analyses build the name of each variable with a plain string join, and no longer with Printf and Format. On 6 open-source repos, the scan time was a median of 3% lower, with identical findings. (dataflow-variable-keys)
The semgrep/semgrep Docker image no longer includes the OpenSSH server (sshd) or sftp-server. It now installs only the OpenSSH client (ssh, scp, ssh-keyscan, ssh-agent, ssh-add and ssh-keygen), which is all git needs to fetch over SSH. Container scanners will no longer report sshd-only CVEs against the image. (docker-ssh-client-only)
go grammar update (v0.25.0):
(go-v0.25.0)
julia grammar update (v0.25.0):
(julia-v0.25.0)
let x = (r := 1; source ()) in sink x,
the value of the sequence is the result of source (), so the finding on
sink x is now reported. Taint also no longer leaks from one match case
into the next one. (ocaml-expression-statements)HTTPS_PROXY=http://user:password@proxy:8080) when it reports the configured
proxy settings. (proxy-log-redact-userinfo)include or a remapped
projectDir in settings.gradle). The resolver's init script reports each
project's actual build file, and dependencies are attributed to that file
instead of falling back to the root manifest. (SC-4327)errors of the JSON output (one entry per resolver error or lockfile parse error of each failed subproject, or a generic entry if none was reported; resolver errors include the failed command and the end of its output, with common credential shapes such as URL credentials and Authorization headers redacted), not only in the log. Previously the JSON had no error, so pipelines that read it could not tell why the scan failed. (SC-4348)Your coding agent can read these notes before it upgrades. Set up the MCP server →
PHP: Improved virtual call resolution for basic trait composition. (CODE-9176)
requirements.txt dependency declared as a version range (litellm>=1.10) or with no version (flask) now produces a finding for every advisory whose affected versions overlap the declared range, in both the Python and native resolvers. Such a finding has no single installed version to confirm, so it needs review. A range line next to a pinned copy of the same package produces no finding of its own, since the pinned copy names the installed version. (SC-4243)SEMGREP_LOG_JSON_FILE to emit structured execution events with scan metadata alongside existing text output. (SMS-3670)Scans are slightly faster. The dataflow analyses build the name of each variable with a plain string join, and no longer with Printf and Format. On 6 open-source repos, the scan time was a median of 3% lower, with identical findings. (dataflow-variable-keys)
go grammar update (v0.25.0):
(go-v0.25.0)
HTTPS_PROXY=http://user:password@proxy:8080) when it reports the configuredinclude or a remappedprojectDir in settings.gradle). The resolver's init script reports eacherrors of the JSON output (one entry per resolver error or lockfile parse error of each failed subproject, or a generic entry if none was reported; resolver errors include the failed command and the end of its output, with common credential shapes such as URL credentials and Authorization headers redacted), not only in the log. Previously the JSON had no error, so pipelines that read it could not tell why the scan failed. (SC-4348)The --help output of semgrep subcommands now includes a "THIRD-PARTY LICENSES" section linking to the licenses and notices for Semgrep's open source d
--help output of semgrep subcommands now includes a "THIRD-PARTYSemgrep no longer ships binaries for Intel Macs; 1.182 is the last release that supports them. On Intel Macs, use the semgrep/semgrep Docker image; on Apple Silicon, use a native arm64 Python. (drop-intel-macos)
Speed up scans with many rules that share file prefilter conditions by checking each condition once per file. (prefilter-predicate-memo)
semgrep ci --no-suppress-errors now exits with code 2 (instead of 0) when dependency resolution fails for a subproject while local builds are allowed (--allow-local-builds): a build or package manager command fails, a lockfile cannot be parsed at all, or a build succeeds but reports an incomplete dependency list (for example, a registry refused a request). The affected dependency sources are listed in the scan output and the scan no longer reports that it completed successfully. Findings from the other subprojects are still reported. The default behavior, where errors are suppressed, is unchanged, and so is any scan without --allow-local-builds. Semgrep Managed Scanning scans are not affected. (SC-3613)
bash grammar update (v0.25.1):
$(( )), $[ ], (( )), C-style for headers) is now translated into operator trees with bash precedence, so patterns such as $(( $X * 2 )) match structurally. $((...)) is no longer confused with $(...).... before done, fi or }, e.g. while ...; do ... done, and until loops now parse as loops.# (x=foo#bar) are no longer truncated, and arguments after a redirect or heredoc delimiter stay with their command.$[ ], time ( ... ), extglob in case and [[ ]], and if function bodies.(bash-v0.25.1)
clojure grammar update (v0.0.13):
(clojure-v0.0.13)
swift grammar update (0.5.0):
#… expression nodes (not boolean Not) so patterns like if !$X do not match availability checks.(swift-0.5.0)
semgrep mcp now track edited files separately for each Cursor conversation, so one window's stop scan no longer scans or discards another conversation's edits. (mcp-hook-cache-per-conversation)semgrep mcp server's semgrep_scan tool now only scans files inside the client's workspace root, or the server's working directory when the client reports none. Set SEMGREP_MCP_ALLOW_SCAN_OUTSIDE_WORKSPACE=true to scan files outside it. (mcp-scan-workspace-roots)Failure: Failed to unmarshal cached value: Exception during parallelization: Exception.Timeout(...) when agodebug ( ... ) block. Such files used to report no dependencies and a parse error. (SC-4268)[project] dependencies table (Poetry 2.0 and later), in dependency groups or as extras, instead of reporting every package with unknown transitivity. Packages from Poetry dependency groups are now reported as direct. PyPI rule matching now ignores differences in case, -, _ and . in package names, so rules match however a lockfile spells the name. (SC-4278)# yarn lockfile v1. Previously these failed with "Failed to parse yarn.lock at 1:1" and no dependencies were analyzed. (SC-4286)Update certificate trust in Semgrep and its container images to address the eMudhra certificate authority compromise. Disable curl-based telemetry on
Added Objective-C language support ( objc , alpha): tree-sitter based parsing of .m files with semgrep pattern syntax (metavariables, ellipsis in argu
objc, alpha): tree-sitter based parsing of .m files with semgrep pattern syntax (metavariables, ellipsis in argument/statement/message-keyword positions, deep ellipsis), enabling search-mode and taint-mode rules on Objective-C code. Message sends, blocks, control flow, class/method definitions, and Objective-C literals are modeled; remaining constructs parse and degrade gracefully. Objective-C++ (.mm) support will follow separately. (gh-9265)c-sharp-pro grammar update (v0.23.1):
@class, cl\u0061ss, and class (or a\u200Cb and ab) are the same name.℘ (U+2118) and · (U+00B7), and still accepts formatting characters (e.g. soft hyphen) inside identifiers.(c-sharp-pro-v0.23.1)
hcl grammar update (v1.2.0):
(hcl-v1.2.0)
lua grammar update (v0.5.0):
(lua-v0.5.0)
requirements.txt line declared as a version range or bare name is now reported as a direct dependency even when a requirements.in manifest is present. Such a line can only have been written by hand, since pip-compile emits exact pins, so it is a direct declaration regardless of whether the manifest names the package. Exact pins keep using the manifest to decide between direct and transitive. (SC-4242)Update certificate trust in Semgrep and its container images to address the eMudhra certificate authority compromise. Disable curl-based telemetry on
semgrep scan now uses the same asynchronous configuration generation and retries as semgrep ci when fetching platform scan configurations. (CODE-9930)
semgrep scan now uses the same asynchronous configuration generation and retries as semgrep ci when fetching platform scan configurations. (CODE-9930)
Python only: Supply Chain now reports every package declared in a requirements.txt, not only the ones pinned with ==. A requirement with a version range (requests>=2.0, numpy~=1.26, django==2.2.*, requests (>=2.0)) is reported with the declared specifier as its version, and a bare name (flask) is reported with an empty version, so they appear in the dependency inventory and SBOM instead of being dropped with a "not pinned to an exact version" parse error. These dependencies have no definitive installed version, so for now they produce no Supply Chain findings. When there is no requirements.in manifest to consult, a range or bare-name line is reported as a direct dependency, since it was declared by hand rather than written by a resolver; exact pins keep their previous unknown transitivity. Local path and URL lines are no longer reported as parse errors, while a line whose leading token is not a valid package name (requests$foo>=1) is now reported as a parse error instead of being dropped silently. Other ecosystems and other Python manifest formats are unchanged. (SC-3487)
pro: Various performance improvements for historical secrets scanning. (SCRT-983-historical-performance)
go grammar update (v0.23.4):
(go-v0.23.4)
Hack grammar update:
1 + 2, preserving the visitor and body text. Embedded DSL bodies remain opaque rather than being parsed as Hack expressions.(hack-1a7ded90288189746c54861ac144ede97df95081)
git gcPromoted Dart to GA maturity. Pattern matching for Dart now supports type inference for typed declarations (id_type propagation), boolean operator typ
&&/|| as bool), alias-aware import equivalence (a pattern like http.get(...) matches code that imports package:http/http.dart under any local alias), method-chain ellipsis (obj.foo(). ... .bar()), multi-statement patterns (e.g. $V = get(); ...; eval($V);), and class-body ellipsis (class $X { ... }). (dart-ga-types-equivalences)--lang ql and the ql language key in rules, and parsing of .ql and .qll targets, are no longer available. (remove-ql)anyio (4.12.1 to 4.14.2) and urllib3 (2.6.3 to 2.8.0)…pick up a PyJWT release with fixes for known vulnerabilities, and projects that require a newer PyJWT can install Semgrep. ( gh-11953 )
bun.lock lockfiles (text format, lockfileVersion 1) paired with a package.json manifest, including the dependency relationships between packages. The deprecated binary bun.lockb format is still unsupported. (SC-2419)Pro: Stabilize file processing order in inter-file analysis, which was previously arbitrary but deterministic. We believe this is very unlikely to affect findings but there is a theoretical path by which it could in certain rare cases. (stabilize)
c grammar update (v0.24.2):
(c-v0.24.2)
cpp grammar update (v0.23.4):
(cpp-v0.23.4)
Update the html grammar to v0.23.2. (html-v0.23.2)
r grammar update (v1.3.0):
(r-v1.3.0)
sfapex grammar update (v2.3):
(sfapex-v2.3)
except A, B: and except A, B, C: (PEP 758, Python 3.14) and theexcept* forms of both are now parsed as a tuple of exception types, likeexcept (A, B):, rather than as the Python 2 except A as B: or a parseexcept* is now handled by Semgrep's primary Python parser too, soexcept A, e: is no longer read as except A as e:, even for python2.except A, B, C: is still a parse error there. TheEXPOSE ports with TCP/UDP protocols. (gh-11934)pyjwt[crypto]>=2.15.0,<3) instead of~=2.13.0, so installs pick up a PyJWT release with fixes for knownThe bundled tree-sitter C runtime is now 0.26.3. (CODE-9425)
semgrep-core exited on Windows. (windows-hang)Added native Supply Chain support for Bazel workspaces using rules_jvm_external . Semgrep now recognizes a maven_install.json pinned lockfile (version
rules_jvm_external. Semgrep now recognizes a maven_install.json pinned lockfile (versions 0.1.0 and 3, as emitted by rules_jvm_external 3.x through current) paired with a MODULE.bazel (or legacy WORKSPACE / WORKSPACE.bazel) marker as a Maven-ecosystem subproject, and attributes findings to the workspace root rather than the lockfile's directory. Workspace-declared root artifacts are identified via __INPUT_ARTIFACTS_HASH for accurate direct-vs-transitive classification; dependencies are emitted with Unknown transitivity when that field is not available. This is the first milestone of native Bazel coverage; broader ecosystem support (rules_python, rules_go, rules_js) and Bazel-aware reachability attribution follow. (SC-2008)FOOBAR(a+)\1 will skip any file that does notFOOBAR without running the regex. (scrt-979)--x-gradle-module-attribution. Enabling it can change finding IDs because finding paths change; the ID calculation is unchanged. (SC-2560)Speed up semgrep ci filtering when a deployment has many triage-ignored findings. (triage-ignored-performance)
Semgrep no longer crashes with an OCaml stack trace when a proxy environment
variable holds an unusable value. HTTP_PROXY, HTTPS_PROXY, or ALL_PROXY
set to an empty value is now ignored with a warning, and the scan
proceeds without a proxy. A non-empty value that is not a usable proxy URL
now exits with an error message, with any credentials in the URL
redacted, instead of failing inside the HTTP client.
Semgrep also now adds the missing scheme to a proxy URL supplied
without one; https for HTTPS_PROXY and http` otherwise. (ENGINE-2208)
Supply Chain: lockfileless Gradle scans now report a "Resource Inaccessible"
resolution error when a repository refuses a request (for example a 401 from a
private registry), instead of exiting successfully with a silently incomplete
dependency list. (sc-3358)
Nothing published for this version
Homebrew installs of Semgrep are no longer supported on Intel Macs. (ENGINE-2951)
Fixed scans run with --trace that would fail with Stdlib.Effect.Unhandled(Eio__core__Cancel.Get_context) when a secret would fail to validate (CODE-99
--trace that would fail with Stdlib.Effect.Unhandled(Eio__core__Cancel.Get_context) when a secret would fail to validate (CODE-9906)Semgrep is now built with OCaml 5.4 instead of OCaml 5.3. End users should notice no change. (ocaml-5.4)
Diff scans now report which dependency source files were added or modified relative to the merge base. (changed-dependency-sources)
--x-partial-scan-rule-id) now resolve dependencies only for
subprojects whose ecosystem is evaluated by the rules being run. Subprojects in
any other ecosystem are skipped and reported to semgrep-app as skipped
subprojects. (SC-3830)semgrep-core -version now reports the git commit that the binary was built from. (core-version-git-sha)
semgrep-core -version now reports the git commit that the binary was builtno_disk_cache memory policy (--x-mem-policy no_disk_cache) to trade higher memory usage for not caching intermediary scan data structures on disk. (no-disk-cache-memory-policy)layout at, assembly flags, global using-directives, EVM Cancun builtins) and corrected ternary/?: vs member-access precedence. (LANG-207)metavariable-regex or a binding-introducingmetavariable-pattern would emit a duplicate finding on the same range whosea hash $ALG was detected)use with a nested path (e.g. use a::b::C). It was wrongly beingAdded support for the OpenTofu .tofu file extension. Because OpenTofu uses the same HCL grammar as Terraform, .tofu files are now automatically detect
.tofu file extension. Because OpenTofu uses the same HCL grammar as Terraform, .tofu files are now automatically detected and scanned as Terraform, so they are picked up by recursive scans and Terraform rulesets (e.g. p/terraform) with no extra configuration. (ENGINE-2884)!=, case/when expressions with line breaks, and element references that take a block. (LANG-206)semgrep ci has been extended from the last 30 days to the last 90 days, to match the updated usage policy. (contributor-window-90-days)--debug is NOT passed, matching the segfault error output when --debug is passed (engine-segv)Scan Status output when no code rules will run (e.g. aAdded support for the OpenTofu .tofu file extension. Because OpenTofu uses the same HCL grammar as Terraform, .tofu files are now automatically detect
.tofu file extension. Because OpenTofu uses the same HCL grammar as Terraform, .tofu files are now automatically detected and scanned as Terraform, so they are picked up by recursive scans and Terraform rulesets (e.g. p/terraform) with no extra configuration. (ENGINE-2884)semgrep ci has been extended from the last 30 days to the last 90 days, to match the updated usage policy. (contributor-window-90-days)Scan Status output when no code rules will run (e.g. aNothing published for this version
Pro C/C++ scans now skip code inside statically-dead preprocessor branches (for example, #if 0 ... #else ... #endif ). Patterns that would otherwise m
#if 0 ... #else ... #endif). Patterns that would otherwisesemgrep install-semgrep-pro now sends usage metrics so that--metrics off or SEMGREP_SEND_METRICS=off. Metrics payloads also#if 0 filtering now also handles cases where the directive splits a#if 0 void foo(int i) { #else void foo(uint32_t i) { #endif. (engine-994)Fixed a crash at startup (Fatal error: Failed to allocate signal stack for domain 0) when running Semgrep on systems with musl 1.2.6 (e.g. Alpine 3.24) on
recent Intel CPUs whose kernel-reported minimum signal-stack size exceeds musl's
build-time SIGSTKSZ (notably AMX-capable Xeons). (ENGINE-2863)
Dockerfile: Fixed parse errors on RUN instructions that use heredoc syntax
(<<EOF, <<-EOF, quoted delimiters). (LANG-263)
metavariable-type now supports fully qualified type names in languages
where a qualified name in type position parses as an expression (e.g.
Python's types: [a.b.C]) when the metavariable's type is determined by
type inference, such as Pro engine cross-file type resolution. (LANG-583)
Updated the ocaml-tree-sitter-core dependency to the latest main.
parser.c with the tree-sitter version that produced it.Updated Dart parser to a more recent upstream version. (LANG-579)
Aliengrep (generic mode) now uses the maintained libpcre2 10.x regular-expression library instead of the deprecated libpcre 8.x. Matching behavior is…
--x-dependency-paths flag to scan and ci that includes the full dependency path(s) for transitive supply-chain findings in --json and --sarif output. (SC-3547)metavariable-regex and metavariable-comparison (re.match()) runtimes now use the maintained libpcre2 10.x library instead of the deprecated libpcre 8.x. Matching behavior is unchanged. (eval-generic-pcre2)Added support for more operators for folding for constant propagation, including subtraction, division, bit ops, bit shifts, comparisons, and more. (c
nosemgrep_disabled field to the scan configuration so the platform can disable nosemgrep inline ignore comments org-wide for a scan. (APPEX-1122)--no-exclude-binary-files tosemgrep ci with --sarif now correctly populates the output's ignoresUpdated the ocaml-tree-sitter-core submodule to the latest upstream main, providing
(ocaml-tree-sitter-core-bump)
Pro: Added experimental cross-file (interfile) analysis for Gosu, enabling taint tracking across multiple Gosu source files. (gosu-interfile)
0x_dead_beef, 0o_755, 0b_1010_1010). (LANG-533)def and class definitions. (LANG-536)~/.semgrep/settings.yml's storedSEMGREP_APP_TOKENsemgrep ci scans originating from a pre-commit hook will no longer fail withUnable to create '<tmp>/.git/index.lock': Not a directory in certain cases. (engine-2736)--x-no-python-schema-validation is still accepted as a no-op with a deprecation warning, and will be removed in a future release. (x-rule-validation)
--max-match-context-size option to limit the number of characters of source code included as context for each match in the output. This prevents matches in minified files (e.g., minified JavaScript where the entire file is a single line) from producing enormous output Set to 0 for unlimited, which is the default value. (ENGINE-2117)--x-no-python-schema-validation with a value-taking --x-rule-validation=full|core-only|none flag. The default (full) preserves existing Python rule validation behavior; core-only matches the old flag's semantics (disables Python rule validation and uses semgrep-core RPC validation only); none skips both pre-validation passes, surfacing rule errors at scan-time. --x-no-python-schema-validation is still accepted as a no-op with a deprecation warning, and will be removed in a future release. (x-rule-validation)validation_error results on HTTP secret validators (Facebook, Slack, Stripe, Google, Cloudflare, etc.) by retrying transient network failures, mirroring the retry behavior already present for AWS validators. (SCRT-965)Dart: typed metavariables ($X as T) and metavariable-type, metavariable binding inside string interpolations, and function-definition patterns that ma
$X as T) and metavariable-type,
metavariable binding inside string interpolations, and function-definition
patterns that match Dart function definitions. (gh-11678)>=2.35 to >=2.34, allowing users on distros
that ship glibc 2.34 (e.g RHEL 9 & AL2023) to install the semgrep wheel. (gh-11622)Baseline diff scans (semgrep ci and --baseline-commit) no longer treat every finding on a file as newly introduced when rule(s) failed during the baseline run.
Per-rule failures (for example a timeout for a single rule) on baseline analysis now hide only that rule's matches on that file from the "new vs baseline" comparison. Other rules on the same file are still taken in comparison for the "new vs baseline" comparison.
Per-file, rule-independent failures now hide all findings on that file from the "new vs baseline" comparison. (LANG-515)
Fixed a yarn.lock parse error on Yarn Berry entries written in YAML explicit-key form. Affected lockfiles previously failed to parse. (SC-3479)
The (beta) SBT resolver with --allow-local-builds now correctly identifies dependencies as part of the Maven ecosystem. (SC-3522)
Fix --sarif-output and --sarif causing nosemgrep-suppressed findings to be reported in CLI scan output and to block scans. Suppressed findings are now correctly excluded from terminal text output, the scan-summary count, and the CLI's exit code. (engine-1824)
Fixed a bug that could cause unreliable target filtering in parallel scans. (gh-6313)
Dart: improved parser fidelity for Dart 3 grammar features and routed
pattern parsing for statements beginning with await, rethrow, and other
statement keywords. Eliminates a large class of PartialParsing errors on
real-world pub.dev packages. (gh-11678)
semgrep-core-proprietary so the binary works when semgrep install-semgrep-pro is invoked, and semgrep is installed via Homebrew. (pro-binary-homebrew)<case>.named_ast.expect golden files for tests/intrafile/maturity/ fixtures, exercised by Unit_maturity_named_asts. (LANG-287)Updated PHP target parsing to support grammar changes from PHP 8.1-8.5 (LANG-380)
semgrep ci startup time with App-provided rules by avoiding duplicate semgrep-core rule validation during CLI rule loading while preserving config-style failures for invalid rules. (ci-rule-validation-startup)json.dumps([rule.raw])) but was being created with a .yaml suffix.
OCaml's Parse_rule.parse_file dispatches purely on file extension, so this
routed every TR rule through Yaml_to_generic.parse_yaml_file (the slow YAML
path) instead of Fast_json.parse_program (the new hand-written RFC 8259
parser). Switching the suffix to .json lines the suffix up with the actual
content and lets every TR rule parse take the fast path. (tr-json-suffix)pro: Improved support for tracking taint through nested functions. (LANG-95)
semgrep_findings tool: added a refs parameter to filter findings by branch (defaults to the primary branch when not specified), and made autotriage_verdict optional so that findings without an AI verdict can also be returned. (engine-2723)import and importstr now reject paths that resolve outside the
rule file's parent directory. (ENGINE-2727)Authorization header
values from git error messages and from the captured tracebacks sent to
the fail-open telemetry endpoint, preventing leaks of secrets like
CI_JOB_TOKEN from a failed git fetch in GitLab CI. Also closes
ENGINE-2731 (raw, unsanitized tracebacks in fail-open telemetry). (ENGINE-2728)semgrep ci no longer transmits SCM tokens to the Semgrep Platform. (ENGINE-2729)~/.semgrep/semgrep.log or $SEMGREP_LOG_FILE) now respects the requested log level instead of always being written at DEBUG. This narrows the surface for credentials to land on disk via CI runner filesystems or job artifacts; pass --debug to restore the previous behavior. (ENGINE-2730)imports
or runtime function calls that recurse forever. (ENGINE-2727-dos)Scala 3.4+ trait parameters are now parsed correctly. (lang-73)
SEMGREP_LOG_SRCS=cohttp.client. (ENGINE-2712)Scala: Added tree-sitter parser for improved parsing accuracy with pfff fallback. (LANG-255)
Semgrep now reports an error instead of silently returning zero findings when target file discovery fails (e.g., due to a git ls-files failure). (ENGI
Added support for a supply chain hook for the Semgrep Plugin (supply-chain-hook)
--no-x-run-taint-once as a flag. (engine-2468)SEMGREP_DISABLE_CONFIG_DOWNLOAD_V2=1 to fall back to the legacy endpoint. (SMS-2284)codeFlows. (engine-2570)pro: Improved taint tracking through lambda calls. (LANG-268)
$C.getInstance(...), and then
use metavariable-type on $C to check its type. (LANG-271)metavariable-type. (LANG-271)The Kotlin tree-sitter parser has been updated to the latest available grammar significantly improving Kotlin support in Semgrep. (kotlin-parser)
semgrep ci when run in a git repo with no remote origin set (gh-11342)Added support for (agentic) hooks in Windsurf. (windsurf-hooks)
Removed the experimental and undocumented command semgrep install-ci. (osemgrep-install-ci)
Migrate from publishing a single Linux wheel with the platform tag musllinux_1_0_<arch>.manylinux2014_<arch> to publishing two separate wheels:
(pypi-linux-tag)
--secrets-timeout flag. (engine-2593)Fix crash on Windows when running semgrep ci with --debug and no blocking findings. The Windows subprocess path incorrectly raised an exception for al
semgrep ci with --debug and no blocking findings. The Windows subprocess path incorrectly raised an exception for all pysemgrep exit codes (including 0), which was silently swallowed in normal mode but propagated as a fatal error when --debug was active. (ENGINE-2491)-j) (engine-2512)semgrep scan $(git ls-files '*.py')) caused one semgrep-core
subprocess to be spawned per file. Roots that are not directories are now
handled directly in Python without any subprocess overhead. (gh-11404)Nothing published for this version
Semgrep core is now optimized with flambda (flambda)
for-yield (LANG-193)def f (a: t) =
foo()
bar()
``` (lang-194)
Hooks (for both Claude Code and Cursor) now pull custom rules from the registry (custom-rules-hooks)
Hooks (for both Claude Code and Cursor) now pull custom rules from the registry (custom-rules-hooks)
Turned on DNS rebinding protection for the MCP server (dns-check)
Environment variables can now be passed to third-party package managers invoked as part of --allow-local-builds dependency resolution via the environment variable SEMGREP_LOCAL_BUILD_ENV, which accepts a JSON object with string keys and string values. (SC-3163)
Memory management policies
A memory policy defines how OCaml's garbage collector should be configured for
a scan. There are two initial policies: "aggressive", the current behaviour,
which trades longer scan times for lower memory use, and "balanced", which
finds a middle ground between reclaiming heap memory in short order while
limiting how often the garbage collector runs. The policy can be configured
via the --x-mem-policy CLI flag for the pro engine; this flag is unused in
the OSS engine. (engine-2055)
Added experimental support for the OpenFGA authorization language. Thanks to Alex Useche (@hex0punk) for the contribution! (gh-11347)
Allows case insensitive string comparisons using lower() and upper() like this:
- metavariable-comparison:
metavariable: $VALUE
comparison: upper(str($VALUE)) == "SEMGREP"
(gh-11502)
Blocking findings that are outputted in the CI output are now labelled as such. (#4394)
Added progress indicators for symbol analysis calculation and upload during CI scans (sc-3103)
glom to at least version 23.3, which includes a fix to a SyntaxWarning
warning log. (gh-11460)Connecting to the Semgrep MCP server via streamableHttp now requires OAuth. (saf-2453)
pipenv to uv for ./cli package management (uv)-alpha in 1.2.3-alpha. (sc-3001)Added a warning in --debug mode when a user runs a parallel scan with a larger value for -j/--jobs than the number of CPUs we detect the host has made
Target objects. Performance should improve on
large repo scans proportionally to the number of files in the repo. (gh-5407)semgrep ci no longer applies autofixes to disk, even when the "Suggest autofixes" toggle in the app is enabled. (saf-2446)Performance: subproject discovery in Supply Chain scans is no longer significantly slowed down by the presence of Git-untracked files resulting in fas
Gradle lockfiles of the form gradle*.lockfile are now supported. Previously, only lockfiles named exactly gradle.lockfile were supported. (SC-2999)
gradle*.lockfile are now supported. Previously, only lockfiles named exactly gradle.lockfile were supported. (SC-2999)semgrep login now supports a --force flag, which ignores existing tokens and starts a new login session. The MCP setup workflow has been updated to use --force too. (saf-2392)Added support for Cursor post-code-generation hooks via new record-file-edit and stop-cli-scan semgrep mcp flags (cursor-hooks)
record-file-edit and stop-cli-scan semgrep mcp flags (cursor-hooks)skipped_paths field to CI scan results to report files that failed to scan due to errors (timeout, OOM, etc.), preventing the app from incorrectly marking findings in those files as fixed (gh-5122)semgrep ci. (sc-2927)mcp python-sdk from 1.16.0 to 1.23.3 (mcp-version)semgrep ci regardless of app settings is now possible with
--x-enable-transitive-reachability (or --x-tr)
and --x-disable-transitive-reachability. (tr-flags)No significant changes.
No significant changes.
No significant changes.
No significant changes.
Added optional user-prompting for classifying findings as true/false positives via MCP Elicitation in the MCP server (behind SEMGREP_FINDINGS_ELICITAT
let ... in expressions in OCaml is now reported. Previously, the location of the let was omitted. This is mainly relevant for autofix. (ocaml-let)Fix issue that could lead to validation failures for certain well-formed rules, such as those with emoji in their messages. (incid-293)
pro: interfile scans no longer default to -j 1; instead, the number of available CPUs on the system is polled as part of a heuristic to determine how
No significant changes.
No significant changes.
Fix issue that could lead to validation failures for certain well-formed rules, such as those with emoji in their messages. (incid-293)
Semgrep will no longer rarely crash when --trace is passed. (incid-280)
Users can opt into the legacy method with the --x-parmap CLI flag, and this deprecates the --x-eio flag (since it is now the default behaviour). (saf-…
--x-parmap CLI flag, and this deprecates the --x-eio
flag (since it is now the default behaviour). (saf-2271)-k/ --hook flag to enable Semgrep scans via Claude Code Agent post-tool hooks (saf-2279)semgrep scan or semgrep ci, the progress bar now always ends at 100%. (SAF-2079)type declaration. (gh-11283)Nothing published for this version
Nothing published for this version
Nothing published for this version
Pro: improved taint handling of match expressions in Scala. In examples like `scala val x = taint match { case Some(t) => t case None => return "examp
match expressions in Scala. In examples likeval x = taint match {
case Some(t) => t
case None => return "example"
}
dataflow should now track taint from taint to x. (code-9085)case $M -> ... :? ... +& test +& ... => ... patterns. (code-9131)--allow-local-builds is passed. (SC-2899)Rule parsing in 1.139.0 was switched to happen solely in semgrep-core. This caused some users to exit with code 7, so this change has been reverted. (
One column per quarter.