.coderabbit.yaml

# SPDX-FileCopyrightText: Copyright (c) 2026, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# 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.

# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
# Docs: https://docs.coderabbit.ai/reference/configuration

language: "en-US"
tone_instructions: >-
  Review this security-critical vulnerability analysis and VEX pipeline.
  Prioritize correctness, trust boundaries, evidence provenance, fail-safe
  behavior, and stable contracts. Be concise; skip lint/style nits and flag
  missing failure-path tests.

reviews:
  profile: assertive
  review_status: true
  review_details: true
  collapse_walkthrough: false
  pre_merge_checks:
    title:
      mode: error
      requirements: >-
        Title must follow Conventional Commits format:
        type(optional-scope)[!]: concise imperative summary.
        Allowed types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert.
        Use lowercase type and scope. If the change is a breaking change, add a ! before the colon.
        Keep it under 72 characters and do not use a trailing period.
    description:
      mode: warning
    issue_assessment:
      mode: warning
    docstrings:
      mode: "off"
  auto_title_instructions: >-
    Generate the PR title in Conventional Commits format:
    type(optional-scope)[!]: concise imperative summary.
    Allowed types: feat, fix, docs, style, refactor, perf, test, build, ci, chore, revert.
    Use lowercase type and scope, add a ! before the colon for breaking changes,
    keep it under 72 characters, and do not use a trailing period.
  auto_review:
    base_branches: ["develop"]
    enabled: true
    drafts: false
    auto_incremental_review: true
    auto_pause_after_reviewed_commits: 0
  path_filters:
    - "!LICENSE"
    - "!aiq/**"
    - "!uv.lock"
    - "!images/*.png"
    - "!src/vuln_analysis/data/eval_datasets/**"
    - "!tests/server/webhooks/fixtures/**"
    - "!**/.gitkeep"
  path_instructions:
    - path: "src/vuln_analysis/pipeline/**"
      instructions: >-
        Pipeline orchestration fans out per vulnerability and shares image-level
        analysis. Check async cancellation, retry classification, concurrency
        limits, partial-result aggregation, and stage transitions. A failed or
        inconclusive analysis must remain unknown/uncertain; never convert
        missing evidence, timeouts, or tool/model errors into not_affected.
        Cache keys must include every input, model, prompt, policy, and image
        identity element that can change a result.
    - path: "src/vuln_analysis/{functions,prompts,plugins}/**"
      instructions: >-
        Agent workflows, prompts, and VEX decisions are security-sensitive.
        Check prompt-injection resistance, strict tool separation, untrusted
        model-output validation, evidence/citation preservation, deterministic
        decision precedence, and bounded retry/token behavior. Container or
        proprietary filesystem content must not leak to web-enabled agents.
        Classification changes require focused tests and must preserve
        uncertainty when the available evidence is insufficient.
    - path: "src/vuln_analysis/tools/**"
      instructions: >-
        Tools consume attacker-controlled identifiers, URLs, package metadata,
        and extracted container filesystems. Scrutinize path traversal and
        symlink escape, SSRF and unsafe redirects, command/argument injection,
        shell use, decompression/resource exhaustion, response-size limits, and
        timeouts. Filesystem tools must stay within the authorized rootfs and
        network tools must not receive private container content.
    - path: "src/vuln_analysis/server/auth.py"
      instructions: >-
        Authentication boundary. Verify JWT issuer, audience, signature,
        algorithm, expiry, and required-claim validation; reject insecure
        defaults and fail closed. Never log bearer tokens, keys, or credentials.
    - path: "src/vuln_analysis/server/webhooks/**"
      instructions: >-
        Outbound webhook security boundary. Check destination validation and
        SSRF resistance, signing and canonicalization, secret handling,
        replay/idempotency behavior, retry bounds, delivery state transitions,
        and redaction from logs and errors.
    - path: "src/vuln_analysis/server/**"
      instructions: >-
        Public service and asynchronous-job API. Review authorization coverage,
        request validation, report-path containment, information disclosure,
        race conditions, cancellation, resource limits, and stable response
        contracts. Errors should be actionable without exposing secrets or host
        filesystem details.
    - path: "src/vuln_analysis/data_models/**"
      instructions: >-
        These models define user-facing workflow, report, and VEX contracts.
        Flag silently relaxed validation, ambiguous defaults, lossy
        serialization, or breaking schema changes. Keep vulnerability IDs,
        image identities, status values, and output paths normalized and
        unambiguous.
    - path: "src/vuln_analysis/utils/{image_extraction,image_config,pre_triage,report_*,vex_details}.py"
      instructions: >-
        Security-sensitive image materialization and evidence processing. Check
        immutable digest handling, archive and path traversal defenses, command
        arguments, rootfs isolation, provenance metadata, parsing of untrusted
        reports, and fail-safe pre-triage/VEX conclusions.
    - path: "src/vuln_analysis/configs/**"
      instructions: >-
        Runtime and model configuration. Check that component names and options
        match registered implementations, secrets are referenced through
        environment variables rather than committed, security controls remain
        enabled by default, and prompt/model/cache namespace changes stay in
        sync. Flag unbounded concurrency, retries, timeouts, or token budgets.
    - path: "deploy/**"
      instructions: >-
        Deployment assets and executable notebooks. Check for secrets, unsafe
        interpolation, untrusted shell input, insecure exposed ports, weak auth
        or TLS defaults, excessive privileges/capabilities, writable or broad
        mounts, missing resource limits, unsafe proxying, and unpinned images.
        Compose changes must preserve supported startup, health, persistence,
        and ownership behavior. Notebook instructions and code must agree with
        the deployment files and current API.
    - path: "**/*.sh"
      instructions: >-
        Shell scripts should quote expansions, use safe failure handling (set
        -euo pipefail where appropriate), avoid eval and shell interpolation of
        untrusted data, preserve exit status, and never print credentials.
    - path: "tests/**"
      instructions: >-
        Tests should be deterministic, isolated from live network/model services,
        and cover failure and adversarial paths as well as success. Security and
        behavior changes need regression tests. Do not weaken assertions or
        broadly mock away the boundary under test merely to make a test pass.
    - path: "{README.md,docs/**,deploy/README.md,quick_start/**}"
      instructions: >-
        User-facing documentation and examples must match current commands,
        configuration names, API schemas, authentication requirements, output
        semantics, and supported deployment flow. Treat security and VEX claims
        as consequential; generated reports support analyst review and must not
        be presented as unquestionable findings.
    - path: ".gitmodules"
      instructions: >-
        The aiq directory is a Git submodule. Review URL and branch changes for
        supply-chain risk. Do not suggest editing files beneath aiq from this
        repository; dependency updates should change the recorded submodule
        commit deliberately.
    - path: "**/*"
      instructions: >-
        Every hand-authored source and configuration file must carry the
        repository's Apache-2.0 SPDX copyright and license header. When a file
        is modified, confirm its copyright range includes the current year.
        Generated artifacts, binary images, data fixtures, and the aiq
        submodule are exempt.
  poem: false
  sequence_diagrams: true
  suggested_reviewers: false
  finishing_touches:
    docstrings:
      enabled: false
    unit_tests:
      enabled: false
  auto_apply_labels: true
  labeling_instructions:
    - label: DO NOT MERGE
      instructions: >-
        Apply this label to non-draft PRs with critical issues, broken security
        boundaries, unsafe VEX conclusions, incomplete work, or other blocking
        concerns.
    - label: duplicate
      instructions: >-
        Apply this label when the PR duplicates another existing PR or issue.
    - label: invalid
      instructions: >-
        Apply this label when the changes are fundamentally invalid or do not
        follow the project's contribution and security requirements.

knowledge_base:
  code_guidelines:
    enabled: true
    filePatterns:
      - ".agents/**/*.md"
      - "CONTRIBUTING.md"