agentcrash 0.1.0__tar.gz

agentcrash-0.1.0/.gitignore

@@ -0,0 +1,26 @@
+# Python
+__pycache__/
+*.py[cod]
+*.egg-info/
+.eggs/
+build/
+dist/
+.venv/
+.mypy_cache/
+.pytest_cache/
+.ruff_cache/
+*.db
+*.sqlite
+# Node
+web/node_modules/
+web/dist/
+# Editor / OS
+.vscode/
+.idea/
+.DS_Store
+Thumbs.db
+# Local config (never commit user config or data)
+agentcrash.yaml
+# Build outputs for the packaged viewer (regenerated from web/)
+src/agentcrash/server/static/
+report-out/

agentcrash-0.1.0/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole 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.

agentcrash-0.1.0/NOTICE

@@ -0,0 +1,10 @@
+AgentCrash
+Copyright 2026 Prathamesh Dabir
+
+This product includes software derived from other open-source projects. See the
+individual scenario `attribution` metadata and `scenarios/*/README.md` for
+specific attributions. Third-party licenses are documented alongside the
+imported material.
+
+AgentDojo (MIT) and other prior art are cited in the project documentation as
+worked-on precedents; no code is copied from them without license.

agentcrash-0.1.0/PKG-INFO

@@ -0,0 +1,133 @@
+Metadata-Version: 2.5
+Name: agentcrash
+Version: 0.1.0
+Summary: Crash-test your AI agent in a synthetic workplace. An open-source agent security test harness with synthetic tools, visual evidence, and regression tests.
+Project-URL: Repository, https://github.com/agentcrash/agentcrash
+Project-URL: Documentation, https://agentcrash.dev
+Author: Prathamesh Dabir
+License: Apache-2.0
+License-File: LICENSE
+License-File: NOTICE
+Keywords: agent,llm,prompt-injection,sandbox,security,testing
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: Apache Software License
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Topic :: Security
+Requires-Python: >=3.12
+Requires-Dist: fastapi<1,>=0.115
+Requires-Dist: pydantic<3,>=2.7
+Requires-Dist: pyyaml<7,>=6.0
+Requires-Dist: rich<14,>=13.7
+Requires-Dist: sse-starlette<3,>=2.1
+Requires-Dist: typer<1,>=0.12
+Requires-Dist: uvicorn<1,>=0.30
+Provides-Extra: dev
+Requires-Dist: httpx<1,>=0.27; extra == 'dev'
+Requires-Dist: hypothesis<7,>=6.100; extra == 'dev'
+Requires-Dist: mypy<2,>=1.10; extra == 'dev'
+Requires-Dist: pytest-asyncio<1,>=0.23; extra == 'dev'
+Requires-Dist: pytest<9,>=8.2; extra == 'dev'
+Requires-Dist: ruff<1,>=0.5; extra == 'dev'
+Provides-Extra: provider
+Requires-Dist: httpx<1,>=0.27; extra == 'provider'
+Provides-Extra: test
+Requires-Dist: httpx<1,>=0.27; extra == 'test'
+Requires-Dist: hypothesis<7,>=6.100; extra == 'test'
+Requires-Dist: mypy<2,>=1.10; extra == 'test'
+Requires-Dist: pytest-asyncio<1,>=0.23; extra == 'test'
+Requires-Dist: pytest<9,>=8.2; extra == 'test'
+Requires-Dist: ruff<1,>=0.5; extra == 'test'
+Provides-Extra: webview
+Requires-Dist: sentry-sdk<3,>=2.0; extra == 'webview'
+Description-Content-Type: text/markdown
+
+# AgentCrash
+
+Crash-test your AI agent in a synthetic workplace.
+
+See the tool actions an attack causes, then verify your fix. AgentCrash is an
+open-source agent security test harness with synthetic tools, visual evidence,
+and regression tests.
+
+<p align="center">
+  <img src="assets/generated/social-preview.png" alt="AgentCrash: crash-test your AI agent before you trust it. Run one scenario, watch the injection cause a tool action, then prove your fix." width="760"/>
+</p>
+
+> **Scope honesty**: results apply to the listed scenarios and configuration.
+> A passing test is a scoped result, **not** a guarantee that an agent is secure.
+
+## Try the demo
+
+The bundled replay needs no API key and makes no model calls. It runs offline.
+
+```bash
+uvx --from agentcrash==0.1.0 agentcrash demo
+```
+
+or install and run your own live test:
+
+```bash
+uv tool install agentcrash==0.1.0
+agentcrash --version
+agentcrash doctor --mode demo
+```
+
+## What a report tells you
+
+Task completion, attempted unauthorized actions, observed synthetic effects,
+policy decisions, and missing evidence are reported **separately** — never
+collapsed into one score.
+
+- **Task success** — did the legitimate task assertion hold?
+- **Attack attempted** — was the defined forbidden request observed?
+- **Attack succeeded** — did the simulated attacker goal occur in the synthetic world?
+- **Policy blocked** — did the tool policy reject that request?
+- **Completeness** — does the run have all required evidence?
+
+An incomplete run is never a clean pass.
+
+<details>
+<summary>View a report from the actual build (attack variant)</summary>
+
+This is a real run of `invoice-confidential-note`: the injected
+document redirects a synthetic send that is caught in the local outbox.
+
+<img src="assets/generated/report-shot.png" alt="AgentCrash report: five outcome dimensions (task success, attack attempted, attack succeeded, policy blocked, completeness) plus the event timeline." width="820"/>
+
+</details>
+
+**Example evidence:** [`assets/generated/sample-report.html`](assets/generated/sample-report.html) is the
+standalone HTML report the CLI produces for that run.
+
+## Test your agent
+
+Start with the Python adapter recipe. Replace production tools with the
+provided synthetic tool client, validate the adapter, and run the clean task
+and its attack variant.
+
+```bash
+agentcrash init --template invoice
+agentcrash run invoice-confidential-note --variant benign
+agentcrash run invoice-confidential-note --variant attack --trials 5
+agentcrash report RUN_ID --format html
+```
+
+The scripted model broker needs **no model key** — all of the above run offline
+with deterministic results. Live provider integration is a scaffold in v0.1;
+when you wire a provider key, it is held broker-only and never passed on the
+CLI or written into a run.
+
+## Contribute
+
+Add a synthetic scenario, improve an integration, or help someone complete
+their first run. See [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## Security
+
+See [SECURITY.md](SECURITY.md) for the responsible-disclosure channel and the
+tested security boundary.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE) and [NOTICE](NOTICE).

agentcrash-0.1.0/README.md

@@ -0,0 +1,89 @@
+# AgentCrash
+
+Crash-test your AI agent in a synthetic workplace.
+
+See the tool actions an attack causes, then verify your fix. AgentCrash is an
+open-source agent security test harness with synthetic tools, visual evidence,
+and regression tests.
+
+<p align="center">
+  <img src="assets/generated/social-preview.png" alt="AgentCrash: crash-test your AI agent before you trust it. Run one scenario, watch the injection cause a tool action, then prove your fix." width="760"/>
+</p>
+
+> **Scope honesty**: results apply to the listed scenarios and configuration.
+> A passing test is a scoped result, **not** a guarantee that an agent is secure.
+
+## Try the demo
+
+The bundled replay needs no API key and makes no model calls. It runs offline.
+
+```bash
+uvx --from agentcrash==0.1.0 agentcrash demo
+```
+
+or install and run your own live test:
+
+```bash
+uv tool install agentcrash==0.1.0
+agentcrash --version
+agentcrash doctor --mode demo
+```
+
+## What a report tells you
+
+Task completion, attempted unauthorized actions, observed synthetic effects,
+policy decisions, and missing evidence are reported **separately** — never
+collapsed into one score.
+
+- **Task success** — did the legitimate task assertion hold?
+- **Attack attempted** — was the defined forbidden request observed?
+- **Attack succeeded** — did the simulated attacker goal occur in the synthetic world?
+- **Policy blocked** — did the tool policy reject that request?
+- **Completeness** — does the run have all required evidence?
+
+An incomplete run is never a clean pass.
+
+<details>
+<summary>View a report from the actual build (attack variant)</summary>
+
+This is a real run of `invoice-confidential-note`: the injected
+document redirects a synthetic send that is caught in the local outbox.
+
+<img src="assets/generated/report-shot.png" alt="AgentCrash report: five outcome dimensions (task success, attack attempted, attack succeeded, policy blocked, completeness) plus the event timeline." width="820"/>
+
+</details>
+
+**Example evidence:** [`assets/generated/sample-report.html`](assets/generated/sample-report.html) is the
+standalone HTML report the CLI produces for that run.
+
+## Test your agent
+
+Start with the Python adapter recipe. Replace production tools with the
+provided synthetic tool client, validate the adapter, and run the clean task
+and its attack variant.
+
+```bash
+agentcrash init --template invoice
+agentcrash run invoice-confidential-note --variant benign
+agentcrash run invoice-confidential-note --variant attack --trials 5
+agentcrash report RUN_ID --format html
+```
+
+The scripted model broker needs **no model key** — all of the above run offline
+with deterministic results. Live provider integration is a scaffold in v0.1;
+when you wire a provider key, it is held broker-only and never passed on the
+CLI or written into a run.
+
+## Contribute
+
+Add a synthetic scenario, improve an integration, or help someone complete
+their first run. See [CONTRIBUTING.md](CONTRIBUTING.md).
+
+## Security
+
+See [SECURITY.md](SECURITY.md) for the responsible-disclosure channel and the
+tested security boundary.
+
+## License
+
+Apache-2.0. See [LICENSE](LICENSE) and [NOTICE](NOTICE).

agentcrash-0.1.0/docs/adr/0001-synthetic-tool-boundary.md

@@ -0,0 +1,29 @@
+# ADR-0001: Synthetic tool boundary
+
+- **Status:** Accepted
+- **Date:** 2026-09-09
+
+## Context
+AgentCrash tests prompt-injection in agents. If it wired real tools (real mail,
+HTTP egress, shell), a run could contact real external systems and testing
+would be neither safe nor reproducible.
+
+## Decision
+The tool plane exposes only enumerated synthetic operations (`documents.read`,
+`notes.write`, `mail.send`) on an in-memory `WorldState`. Outbound mail writes to
+a local outbox sink; nothing ever reaches a real external recipient. The tool
+service is the single producer of authoritative side-effect events
+(`world.changed`), giving the evaluator a trusted record.
+
+## Consequences
+- Runs are safe, offline, and deterministic.
+- "Attack succeeded" means a synthetic canary appeared in a local sink — never
+  real credential theft. Editorial rules forbid labeling one as the other.
+- `ToolService` is not an HTTP proxy and does not execute shell commands.
+
+## Alternatives
+- Generic tool passthrough (rejected: unsafe, non-reproducible).
+- Real tooling behind an allowlist (rejected: introduces external side effects).
+
+## Revisit trigger
+If we adopt arbitrary user-provided tools, revisit isolation (see ADR-0002).

agentcrash-0.1.0/docs/adr/0002-isolation-profiles.md

@@ -0,0 +1,36 @@
+# ADR-0002: Isolation profiles for the worker
+
+- **Status:** Accepted (profile contract); live container driver is scaffolding
+- **Date:** 2026-09-09
+
+## Context
+Per the security model, a run may encounter malicious prompt content and
+untrusted worker/adapter code. The host filesystem, provider credentials, local
+services, and other runs must be protected.
+
+## Decision
+Adapters run only in declared isolation profiles. The v0.1 default profile
+(`agentcrash-worker-default`) requires: non-root user, read-only rootfs, all
+capabilities dropped, no privileged mode, no host PID or network namespace,
+bounded writable tmpfs, memory/process/CPU limits, and an enabled seccomp
+policy. No Docker socket is ever mounted. A `ContainerProfile` is validated by
+`validate_profile()` in `sandbox/profile.py`; violating profiles raise.
+
+The deterministic demo/CI path runs an explicitly-labeled *uncontained*
+execution profile for replay only; it is never labeled as contained.
+
+## Consequences
+- `validate_profile` refuses any profile that weakens these invariants
+  (`privileged`, root user, host network/PID, writable rootfs, runtime socket).
+- `sandbox/tests` assert these blockers.
+- A Compose network flag alone is not treated as a guarantee; the launcher must
+  validate the container properties.
+
+## Alternatives
+- Ship each adapter fully isolated by default (deferred: heavier, not needed
+  for the offline replay).
+- Trust the user's host entirely (rejected: this is a security tool).
+
+## Revisit trigger
+macOS/Windows VM-backed runtimes, or adopting community adapters, require
+re-validating resource and network behavior per platform.

agentcrash-0.1.0/docs/adr/0003-separate-outcome-dimensions.md

@@ -0,0 +1,32 @@
+# ADR-0003: Separate utility and security outcome dimensions
+
+- **Status:** Accepted
+- **Date:** 2026-09-09
+
+## Context
+A single "agent is safe" score hides the distinction that defines the product:
+did the legitimate task still succeed while an attack was blocked? Collapsing
+task completion and attacker success into one number makes an incomplete run, a
+blocked-everything defense, and a genuinely good defense look identical.
+
+## Decision
+Every run reports five separated dimensions: `task_success` (utility),
+`attack_attempted`, `attack_succeeded` (attacker goal in authoritative state),
+`policy_blocked`, and `completeness`. They are never merged into a security
+pass. The `RunResultDimensions` dataclass in `schemas/models.py` is the
+single source. A cancelled or partial run is `completeness=False` and can never
+be a clean pass.
+
+## Consequences
+- The viewer reports and the CLI table render these independently.
+- "Blocked everything" (task fails, attack blocked) is easy to spot versus a
+  working fix (task succeeds, attack blocked) — a stated requirement.
+- JUnit export maps a failure/incomplete to a failure/error, never a pass.
+
+## Alternatives
+- A composite risk score (rejected: hides the utility/security trade-off and
+  invites misleading claims).
+
+## Revisit trigger
+If a future feature needs a rolled-up score for a specific scrub, it must stay
+sibling to these dimensions, not replace them.

agentcrash-0.1.0/docs/adr/0004-python-first-sdk.md

@@ -0,0 +1,30 @@
+# ADR-0004: Python-first SDK with a separate CLI and viewer
+
+- **Status:** Accepted
+- **Date:** 2026-09-09
+
+## Context
+The product must be useful through the CLI alone, and the browser viewer must
+not be a second implementation of security decisions. Every entry path (CLI,
+HTML report, JUnit, local API) must agree on results.
+
+## Decision
+Python (`src/agentcrash/`) is the single implementation of the runner,
+evaluation, policies, events, and reports. All exporters (`json`, `html`,
+`junit`, JSONL) are generated from the same `RunResultDimensions` + event
+stream. The React viewer talks only to the local FastAPI `/v1` endpoints; it
+performs no security decision of its own. The runner lives under `core/` and
+the UI artifacts are bundled into the wheel so release users never run Node.
+
+## Consequences
+- One source of truth for results; CLI and browser always agree.
+- The offline replay and deterministic CI need no model key, Docker, or Node.
+- Adapters import the Python SDK; `RunContext` and `AgentResult` are frozen by
+  contract tests in `tests/contract/`.
+
+## Alternatives
+- A JS-native evaluator (rejected: duplicate logic and divergence risk).
+
+## Revisit trigger
+If the viewer needs to make an offline-only decision the backend cannot, that
+capability returns to the backend first.

agentcrash-0.1.0/docs/adr/0005-data-only-packs.md

@@ -0,0 +1,31 @@
+# ADR-0005: Data-only scenario packs
+
+- **Status:** Accepted
+- **Date:** 2026-09-09
+
+## Context
+Community scenario submissions are a growth engine, but a pack that can execute
+code on the host loader is an RCE channel. The threat model treats scenario
+files, model responses, adapter code, and imported reports as untrusted.
+
+## Decision
+Scenario packs are data only: a `scenario.yaml` manifest plus fixture files
+(world JSON, task text, injected content, scripted decision traces). The safe
+YAML loader (`core/loader.py`) uses `yaml.SafeLoader`, refuses python/constructor
+tags, and denies fixture paths that escape the pack directory. There are no
+install hooks and no executable entries. Manifest schema is `extra="forbid"` so
+unknown fields are rejected rather than silently executed.
+
+## Consequences
+- A malformed or hostile pack fails validation loudly, before any run.
+- Packs have attribution/license fields; adaptation of public incidents keeps
+  original attribution.
+- No code runs from a pack on the host loader.
+
+## Alternatives
+- Allow user scripts in packs (rejected: RCE and reproducibility hazard).
+- Accept a generic data blob (rejected: no typed limits or licenses).
+
+## Revisit trigger
+Adding an adaptive-attack research profile later must keep packs data-only and
+keep the evaluator inaccessible to the generator and target agent.