switch-trust-cli 1.7.2__tar.gz → 1.8.0__tar.gz

{switch_trust_cli-1.7.2 → switch_trust_cli-1.8.0}/LICENSE

@@ -182,29 +182,3 @@ responsibility, not on behalf of any other Contributor, and only if You agree to
 indemnify, defend, and hold each Contributor harmless for any liability incurred
 by, or claims asserted against, such Contributor by reason of your accepting any
 such warranty or additional liability.
-
-END OF TERMS AND CONDITIONS
-
-
-APPENDIX: How to apply the Apache License to your work
-
-To apply the Apache License to your work, attach the following boilerplate
-notice, with the fields enclosed by brackets "[]" replaced with your own
-identifying information. (Don't include the brackets!) The text should be
-enclosed in the appropriate comment syntax for the file format. We also
-recommend that a file or class name and description of purpose be included on
-the same "printed page" as the copyright notice for easier identification
-within third-party archives.
-
-Copyright [yyyy] [name of copyright owner]
-
-Licensed under the Apache License, Version 2.0 (the "License"); you may not use
-this file except in compliance with the License. You may obtain a copy of the
-License at
-
-http://www.apache.org/licenses/LICENSE-2.0
-
-Unless required by applicable law or agreed to in writing, software distributed
-under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
-CONDITIONS OF ANY KIND, either express or implied. See the License for the
-specific language governing permissions and limitations under the License.

{switch_trust_cli-1.7.2/switch_trust_cli.egg-info → switch_trust_cli-1.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: switch-trust-cli
-Version: 1.7.2
+Version: 1.8.0
 Summary: Command-line tool for evaluating agents, offering two features to assess agents via code scanning (`switch-trust scan`, whitebox testing) and runtime evaluation (`switch-trust eval`, blackbox testing)
 Author-email: SandboxAQ <info@switchagents.ai>
 License: Software: Switch Trust
@@ -188,60 +188,36 @@ License: Software: Switch Trust
         by, or claims asserted against, such Contributor by reason of your accepting any
         such warranty or additional liability.
         
-        END OF TERMS AND CONDITIONS
-        
-        
-        APPENDIX: How to apply the Apache License to your work
-        
-        To apply the Apache License to your work, attach the following boilerplate
-        notice, with the fields enclosed by brackets "[]" replaced with your own
-        identifying information. (Don't include the brackets!) The text should be
-        enclosed in the appropriate comment syntax for the file format. We also
-        recommend that a file or class name and description of purpose be included on
-        the same "printed page" as the copyright notice for easier identification
-        within third-party archives.
-        
-        Copyright [yyyy] [name of copyright owner]
-        
-        Licensed under the Apache License, Version 2.0 (the "License"); you may not use
-        this file except in compliance with the License. You may obtain a copy of the
-        License at
-        
-        http://www.apache.org/licenses/LICENSE-2.0
-        
-        Unless required by applicable law or agreed to in writing, software distributed
-        under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
-        CONDITIONS OF ANY KIND, either express or implied. See the License for the
-        specific language governing permissions and limitations under the License.
-        
 Project-URL: Homepage, https://github.com/sandbox-quantum/switch-trust-cli
 Project-URL: Issues, https://github.com/sandbox-quantum/switch-trust-cli/issues
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown
 License-File: LICENSE
 Requires-Dist: aiohttp>=3.14.3
-Requires-Dist: anthropic==0.87.0
+Requires-Dist: anthropic>=0.25.0
 Requires-Dist: pyOpenSSL>=26.4.0
 Requires-Dist: cryptography>=50.0.0
 Requires-Dist: bandit==1.9.4
 Requires-Dist: cvss>=2.6
-Requires-Dist: datafog>=4.4.0
+Requires-Dist: datafog==4.4.0
 Requires-Dist: dataclasses-json==0.6.7
 Requires-Dist: detect-secrets==1.5.0
-Requires-Dist: google-adk==2.2.0
-Requires-Dist: litellm>=1.83.7
-Requires-Dist: openai==2.29.0
-Requires-Dist: opentelemetry-api<=1.41.1,>=1.38.0
-Requires-Dist: opentelemetry-exporter-otlp-proto-http<=1.41.1,>=1.38.0
-Requires-Dist: opentelemetry-sdk<=1.41.1,>=1.38.0
+Requires-Dist: google-adk==1.38.0
+Requires-Dist: google-auth>=2.60.0
+Requires-Dist: google-genai>=1.0.0
+Requires-Dist: httpx==0.28.1
+Requires-Dist: litellm==1.100.4
+Requires-Dist: openai>=1.0.0
+Requires-Dist: opentelemetry-api==1.38.0
+Requires-Dist: opentelemetry-exporter-otlp-proto-http==1.38.0
+Requires-Dist: opentelemetry-sdk==1.38.0
 Requires-Dist: pip-audit==2.10.1
-Requires-Dist: protobuf<7,>=6.33.5
+Requires-Dist: protobuf==6.33.5
 Requires-Dist: pydantic==2.12.5
-Requires-Dist: python-dotenv==1.2.2
-Requires-Dist: requests>=2.33.0
-Requires-Dist: rich>=13.0
-Requires-Dist: msgpack>=1.2.1
-Requires-Dist: pydantic-settings>=2.14.2
+Requires-Dist: python-dotenv>=1.0.0
+Requires-Dist: PyYAML>=6.0.0
+Requires-Dist: requests==2.34.2
+Requires-Dist: rich>=13.0.0
 Provides-Extra: full
 Requires-Dist: garak==0.14.0; extra == "full"
 Requires-Dist: transformers[torch]==5.10.1; extra == "full"
@@ -415,6 +391,42 @@ switch-trust scan agent.py
 
 # Specify output file
 switch-trust scan /path/to/code --output results.json
+
+# Fail the build on any high or critical finding
+switch-trust scan /path/to/code --fail-on high
+```
+
+#### Gating CI on severity
+
+`--fail-on SEVERITY` turns a scan into a quality gate. Without it, a completed
+scan always exits `0` however many findings it reported.
+
+`SEVERITY` is one of `critical`, `high`, `medium`, `low`, `info`, and the gate
+trips on any finding **at or above** it — `--fail-on high` fires on a `high` as
+well as a `critical`.
+
+Two details worth knowing before you rely on it:
+
+- **It counts findings after triage**, the ones in the summary and the output
+  file. Triage removes false positives, so a finding it downgrades is judged on
+  its new severity.
+- **A finding whose severity cannot be read trips the gate**, whatever the
+  threshold, and the message names the value. A security gate that silently
+  skipped what it could not parse would be worse than no gate.
+
+The report is always written before the gate is checked, so a failing job still
+has its results file to collect.
+
+| Exit code | Meaning |
+|---|---|
+| `0` | no finding reached the threshold |
+| `5` | at least one did, or one had an unreadable severity |
+| `20` | the scanner model's API key is missing or was rejected |
+| `23` | nothing scannable was found at the path |
+
+```bash
+switch-trust scan /path/to/code --fail-on high
+if [ $? -eq 5 ]; then echo "security gate failed"; fi
 ```
 
 [Full scan guide](https://docs.switchagents.ai/switch-trust-cli/scan/getting-started)
@@ -437,6 +449,75 @@ switch-trust eval model-evaluations attach \
 
 # Run all evaluations for an agent
 switch-trust eval run --model my-agent
+
+# Fail the build if any evaluation scores below 0.8
+switch-trust eval run --model my-agent --fail-under 0.8
+```
+
+#### Gating CI on a score
+
+`--fail-under SCORE` turns a run into a quality gate. Without it, a completed run
+always exits `0` whatever it scored.
+
+`SCORE` is between 0 and 1, and it is the bar each evaluation must **meet** — a
+score of exactly `0.8` passes `--fail-under 0.8`. The check is per evaluation, not
+on the average, so a single failing evaluation cannot be averaged away by
+others that passed.
+
+The report is always written before the gate is checked, so a failing job still
+has its results file to collect.
+
+A run covering several evaluations stops at the first one that hits a problem
+re-running cannot fix — a missing judge credential, an unreadable prompt file,
+an uninstalled optional tool — rather than paying for the rest to fail the same
+way. It also gives up once three evaluations fail in a row, on the assumption
+the target is down. Either way, whatever scored before that point is still
+printed and written to the report, and the exit code is the one for the problem
+rather than `5`: a run that did not finish has no verdict for the gate to
+judge.
+
+| Exit code | Meaning |
+|---|---|
+| `0` | every evaluation ran, and every score met the bar |
+| `5` | at least one evaluation scored below it |
+| `6` | the run has no verdict: nothing scored, or it stopped with evaluations unrun |
+| `20` | an API key is missing, or the provider rejected it |
+| `23` | no evaluations matched the model or tag filters |
+
+Exits `6` and `20` are reported whether or not `--fail-under` was passed. A
+credential problem, or a run with no verdict to report, is a broken run rather
+than a verdict — reporting success for it is the false green a gate exists to
+prevent. Only exit `5`, the verdict itself, is opt-in.
+
+Exit `6` covers both ways a run can end up without a verdict, and the message
+says which: no evaluation produced a score at all, or the run gave up partway
+and some of the evaluations you selected were never attempted. The second is
+still exit `6` even when the ones that did run all scored above the bar — a
+gate applied to the part of a suite that happened to run first is not a result
+you can act on.
+
+Exit `20` covers two credentials, and the message says which:
+
+- **The model under test.** A key that is simply *absent* is refused before
+  the run starts, so you pay no CI time to find out; a key the provider
+  *rejects* is reported after the run, since it takes a call to discover.
+  Either way, set the provider's variable (`GEMINI_API_KEY` /
+  `GOOGLE_API_KEY`, `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`) or put a key in the
+  model's config entry.
+- **The judge/generator** named by `GENERATOR_MODEL`, which adversarial probes
+  and LLM-as-judge detectors need in addition to the model under test. For a
+  `gemini` generator set `GOOGLE_API_KEY` or `GEMINI_API_KEY`, or
+  `GOOGLE_VERTEX_CREDENTIALS_JSON` plus `GOOGLE_CLOUD_LOCATION` for Vertex AI.
+
+> **Upgrading:** this is the one behaviour change in the exit codes. Earlier
+> versions exited `0` when every evaluation errored, and when a run gave up
+> after three consecutive failures having scored something first — so a total
+> provider outage was indistinguishable from a clean run. If your pipeline
+> relies on that, branch on `6` explicitly.
+
+```bash
+switch-trust eval run --model my-agent --fail-under 0.8
+if [ $? -eq 5 ]; then echo "quality gate failed"; fi
 ```
 
 The `switch-trust eval` command requires configuration. See [Configuration](https://docs.switchagents.ai/switch-trust-cli/eval/eval-configuration) to:
@@ -454,7 +535,7 @@ The `switch-trust eval` command requires configuration. See [Configuration](http
 - [Configuration](https://docs.switchagents.ai/switch-trust-cli/eval/eval-configuration)
 - [Environment variables](https://docs.switchagents.ai/switch-trust-cli/reference/env-vars)
 - [Built-in evaluations](https://docs.switchagents.ai/switch-trust-cli/reference/builtin-evaluations)
-- [Data privacy](https://docs.switchagents.ai/switch-trust-cli/reference/data-privacy)
+- [Data privacy](https://docs.switchagents.ai/legal/cli-data-privacy)
 - [FAQ](https://docs.switchagents.ai/switch-trust-cli/resources/faq)
 
 ## Data privacy
@@ -463,7 +544,7 @@ Switch Trust runs on your machine, but several features can call external LLM pr
 (located in `~/.switch-trust/.env`, created by `switch-trust init`). You can set this to a remote managed LLM (i.e. `gemini`, `openai`, `anthropic`)
 or a locally hosted LLM (i.e. `litellm` or `ollama`).
 
-[Read more](https://docs.switchagents.ai/switch-trust-cli/reference/data-privacy).
+[Read more](https://docs.switchagents.ai/legal/cli-data-privacy).
 
 ### Telemetry
 

{switch_trust_cli-1.7.2 → switch_trust_cli-1.8.0}/README.md

@@ -166,6 +166,42 @@ switch-trust scan agent.py
 
 # Specify output file
 switch-trust scan /path/to/code --output results.json
+
+# Fail the build on any high or critical finding
+switch-trust scan /path/to/code --fail-on high
+```
+
+#### Gating CI on severity
+
+`--fail-on SEVERITY` turns a scan into a quality gate. Without it, a completed
+scan always exits `0` however many findings it reported.
+
+`SEVERITY` is one of `critical`, `high`, `medium`, `low`, `info`, and the gate
+trips on any finding **at or above** it — `--fail-on high` fires on a `high` as
+well as a `critical`.
+
+Two details worth knowing before you rely on it:
+
+- **It counts findings after triage**, the ones in the summary and the output
+  file. Triage removes false positives, so a finding it downgrades is judged on
+  its new severity.
+- **A finding whose severity cannot be read trips the gate**, whatever the
+  threshold, and the message names the value. A security gate that silently
+  skipped what it could not parse would be worse than no gate.
+
+The report is always written before the gate is checked, so a failing job still
+has its results file to collect.
+
+| Exit code | Meaning |
+|---|---|
+| `0` | no finding reached the threshold |
+| `5` | at least one did, or one had an unreadable severity |
+| `20` | the scanner model's API key is missing or was rejected |
+| `23` | nothing scannable was found at the path |
+
+```bash
+switch-trust scan /path/to/code --fail-on high
+if [ $? -eq 5 ]; then echo "security gate failed"; fi
 ```
 
 [Full scan guide](https://docs.switchagents.ai/switch-trust-cli/scan/getting-started)
@@ -188,6 +224,75 @@ switch-trust eval model-evaluations attach \
 
 # Run all evaluations for an agent
 switch-trust eval run --model my-agent
+
+# Fail the build if any evaluation scores below 0.8
+switch-trust eval run --model my-agent --fail-under 0.8
+```
+
+#### Gating CI on a score
+
+`--fail-under SCORE` turns a run into a quality gate. Without it, a completed run
+always exits `0` whatever it scored.
+
+`SCORE` is between 0 and 1, and it is the bar each evaluation must **meet** — a
+score of exactly `0.8` passes `--fail-under 0.8`. The check is per evaluation, not
+on the average, so a single failing evaluation cannot be averaged away by
+others that passed.
+
+The report is always written before the gate is checked, so a failing job still
+has its results file to collect.
+
+A run covering several evaluations stops at the first one that hits a problem
+re-running cannot fix — a missing judge credential, an unreadable prompt file,
+an uninstalled optional tool — rather than paying for the rest to fail the same
+way. It also gives up once three evaluations fail in a row, on the assumption
+the target is down. Either way, whatever scored before that point is still
+printed and written to the report, and the exit code is the one for the problem
+rather than `5`: a run that did not finish has no verdict for the gate to
+judge.
+
+| Exit code | Meaning |
+|---|---|
+| `0` | every evaluation ran, and every score met the bar |
+| `5` | at least one evaluation scored below it |
+| `6` | the run has no verdict: nothing scored, or it stopped with evaluations unrun |
+| `20` | an API key is missing, or the provider rejected it |
+| `23` | no evaluations matched the model or tag filters |
+
+Exits `6` and `20` are reported whether or not `--fail-under` was passed. A
+credential problem, or a run with no verdict to report, is a broken run rather
+than a verdict — reporting success for it is the false green a gate exists to
+prevent. Only exit `5`, the verdict itself, is opt-in.
+
+Exit `6` covers both ways a run can end up without a verdict, and the message
+says which: no evaluation produced a score at all, or the run gave up partway
+and some of the evaluations you selected were never attempted. The second is
+still exit `6` even when the ones that did run all scored above the bar — a
+gate applied to the part of a suite that happened to run first is not a result
+you can act on.
+
+Exit `20` covers two credentials, and the message says which:
+
+- **The model under test.** A key that is simply *absent* is refused before
+  the run starts, so you pay no CI time to find out; a key the provider
+  *rejects* is reported after the run, since it takes a call to discover.
+  Either way, set the provider's variable (`GEMINI_API_KEY` /
+  `GOOGLE_API_KEY`, `OPENAI_API_KEY`, `ANTHROPIC_API_KEY`) or put a key in the
+  model's config entry.
+- **The judge/generator** named by `GENERATOR_MODEL`, which adversarial probes
+  and LLM-as-judge detectors need in addition to the model under test. For a
+  `gemini` generator set `GOOGLE_API_KEY` or `GEMINI_API_KEY`, or
+  `GOOGLE_VERTEX_CREDENTIALS_JSON` plus `GOOGLE_CLOUD_LOCATION` for Vertex AI.
+
+> **Upgrading:** this is the one behaviour change in the exit codes. Earlier
+> versions exited `0` when every evaluation errored, and when a run gave up
+> after three consecutive failures having scored something first — so a total
+> provider outage was indistinguishable from a clean run. If your pipeline
+> relies on that, branch on `6` explicitly.
+
+```bash
+switch-trust eval run --model my-agent --fail-under 0.8
+if [ $? -eq 5 ]; then echo "quality gate failed"; fi
 ```
 
 The `switch-trust eval` command requires configuration. See [Configuration](https://docs.switchagents.ai/switch-trust-cli/eval/eval-configuration) to:
@@ -205,7 +310,7 @@ The `switch-trust eval` command requires configuration. See [Configuration](http
 - [Configuration](https://docs.switchagents.ai/switch-trust-cli/eval/eval-configuration)
 - [Environment variables](https://docs.switchagents.ai/switch-trust-cli/reference/env-vars)
 - [Built-in evaluations](https://docs.switchagents.ai/switch-trust-cli/reference/builtin-evaluations)
-- [Data privacy](https://docs.switchagents.ai/switch-trust-cli/reference/data-privacy)
+- [Data privacy](https://docs.switchagents.ai/legal/cli-data-privacy)
 - [FAQ](https://docs.switchagents.ai/switch-trust-cli/resources/faq)
 
 ## Data privacy
@@ -214,7 +319,7 @@ Switch Trust runs on your machine, but several features can call external LLM pr
 (located in `~/.switch-trust/.env`, created by `switch-trust init`). You can set this to a remote managed LLM (i.e. `gemini`, `openai`, `anthropic`)
 or a locally hosted LLM (i.e. `litellm` or `ollama`).
 
-[Read more](https://docs.switchagents.ai/switch-trust-cli/reference/data-privacy).
+[Read more](https://docs.switchagents.ai/legal/cli-data-privacy).
 
 ### Telemetry
 

{switch_trust_cli-1.7.2 → switch_trust_cli-1.8.0}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "switch-trust-cli"
-version = "1.7.2"
+version = "1.8.0"
 description = "Command-line tool for evaluating agents, offering two features to assess agents via code scanning (`switch-trust scan`, whitebox testing) and runtime evaluation (`switch-trust eval`, blackbox testing)"
 readme = "README.md"
 license = { file = "LICENSE" }
@@ -8,9 +8,12 @@ requires-python = ">=3.11"
 authors = [
   { name = "SandboxAQ", email = "info@switchagents.ai" },
 ]
+# These versions are kept in step with the versions the library is developed
+# and tested against upstream, so an installed `switch-trust` runs what was
+# tested. Please raise an issue rather than re-pinning one in isolation.
 dependencies = [
   "aiohttp>=3.14.3",
-  "anthropic==0.87.0",
+  "anthropic>=0.25.0",
   # Direct floor pin for transitive dependency of google-adk->google-auth
   # required for `pip-compile` to resolve the dependency tree.
   "pyOpenSSL>=26.4.0",
@@ -19,32 +22,34 @@ dependencies = [
   "cryptography>=50.0.0",
   "bandit==1.9.4",
   "cvss>=2.6",
-  "datafog>=4.4.0",
+  "datafog==4.4.0",
   "dataclasses-json==0.6.7",
   "detect-secrets==1.5.0",
-  "google-adk==2.2.0",
-  "litellm>=1.83.7",
-  "openai==2.29.0",
-  # Cap opentelemetry at 1.41.1 to match google-adk's hard ceiling
-  # (opentelemetry-api/sdk >=1.36,<=1.41.1). Without the cap, pip-compile
-  # grabs the newest otel (1.42+) then backtracks the whole otel stack down
-  # in lockstep, which is slow and can spuriously hit ResolutionImpossible.
-  "opentelemetry-api>=1.38.0,<=1.41.1",
-  "opentelemetry-exporter-otlp-proto-http>=1.38.0,<=1.41.1",
-  "opentelemetry-sdk>=1.38.0,<=1.41.1",
+  "google-adk==1.38.0",
+  # switch_trust/eval/core/models/{generator_model,
+  # model_vertex_agent_runtime}.py build service-account credentials with it.
+  "google-auth>=2.60.0",
+  "google-genai>=1.0.0",
+  # switch_trust/scan/exception_classification.py classifies httpx transport
+  # errors.
+  "httpx==0.28.1",
+  "litellm==1.100.4",
+  "openai>=1.0.0",
+  "opentelemetry-api==1.38.0",
+  "opentelemetry-exporter-otlp-proto-http==1.38.0",
+  "opentelemetry-sdk==1.38.0",
   "pip-audit==2.10.1",
   # google-adk chains the whole otel stack to <=1.42.1, whose
   # opentelemetry-proto still requires protobuf<7.0, while google-api-core /
-  # googleapis-common-protos / grpcio-status require protobuf>=6.33.5. That
-  # leaves 6.33.x as the only viable window; pin it so the resolver lands
-  # there directly instead of backtracking down from protobuf 7.x.
-  "protobuf>=6.33.5,<7",
+  # googleapis-common-protos / grpcio-status require protobuf>=6.33.5. The
+  # exact pin keeps the resolver from backtracking down from protobuf 7.x.
+  "protobuf==6.33.5",
   "pydantic==2.12.5",
-  "python-dotenv==1.2.2",
-  "requests>=2.33.0",
-  "rich>=13.0",
-  "msgpack>=1.2.1",
-  "pydantic-settings>=2.14.2",
+  "python-dotenv>=1.0.0",
+  # switch_trust/scan/static_scanner.py parses the OpenGrep rule file.
+  "PyYAML>=6.0.0",
+  "requests==2.34.2",
+  "rich>=13.0.0",
 ]
 
 [project.optional-dependencies]

switch_trust_cli-1.8.0/switch_trust/cli/__main__.py

@@ -0,0 +1,7 @@
+from switch_trust.cli.main import main
+
+# `raise SystemExit(main())`, not a bare `main()`: the installed console script
+# is generated as `sys.exit(main())`, so dropping the return value here would
+# make `python -m switch_trust.cli` (and the Bazel binary, which runs this same
+# file) exit 0 on a run the installed binary fails. See exit_codes.py.
+raise SystemExit(main())

{switch_trust_cli-1.7.2 → switch_trust_cli-1.8.0}/switch_trust/cli/console.py

@@ -101,6 +101,9 @@ def select(
         while True:
             ch = sys.stdin.read(1)
 
+            if not ch:
+                break
+
             if ch == "\r" or ch == "\n":
                 break