Agent-design-language repo-review-security

Specialist security reviewer for a multi-agent repository review. Use when a review packet needs a bounded AppSec role focused on trust boundaries, secret handling, injection risks, privilege and permission failures, unsafe file/network IO, deserialization, supply-chain exposure, and abuse paths without performing remediation or broad synthesis.

install

source · Clone the upstream repo

git clone https://github.com/danielbaustin/agent-design-language

Claude Code · Install into ~/.claude/skills/

T=$(mktemp -d) && git clone --depth=1 https://github.com/danielbaustin/agent-design-language "$T" && mkdir -p ~/.claude/skills && cp -r "$T/adl/tools/skills/repo-review-security" ~/.claude/skills/danielbaustin-agent-design-language-repo-review-security && rm -rf "$T"

manifest: adl/tools/skills/repo-review-security/SKILL.md

source content

Repo Review Security

Review a repository slice for security and abuse-path findings as one specialist in the multi-agent repo review suite.

This is a review skill, not a threat-modeling replacement and not a remediation workflow. If the operator explicitly asks for a full threat model, use the dedicated threat-modeling path instead.

Quick Start

Confirm the repo, branch, path, diff, or review packet scope.
Identify trust boundaries, secret surfaces, external inputs, filesystem and network effects, authz/authn checks, and dependency/config risks.
Review concrete exploitability before generic hardening advice.
Emit findings first, with severity, affected surface, and abuse scenario.
Hand the artifact to
```
repo-review-synthesis
```
for cross-role assembly.

Focus

Prioritize:

secret leakage, token handling, and credential persistence
injection, command execution, path traversal, and unsafe parsing
privilege, permission, tenancy, and authorization failures
unsafe network, filesystem, deserialization, and artifact handling
supply-chain, dependency, CI, and build configuration exposure
security-relevant logging, diagnostics, and durable artifact leaks

Defer primary ownership of these areas to other specialists:

ordinary correctness and maintainability:
```
repo-review-code
```
test coverage quality:
```
repo-review-tests
```
docs and onboarding truth:
```
repo-review-docs
```
final dedupe and ordering:
```
repo-review-synthesis
```

Required Inputs

At minimum, gather:

```
repo_root
```

one concrete target:

```
target.target_path
```
```
target.branch
```
```
target.diff_base
```
```
target.review_packet_path
```

Useful additional inputs:

```
changed_paths
```
```
trust_boundaries
```
```
sensitive_assets
```
```
exclude_paths
```
```
validation_mode
```

If there is no concrete repo or slice target, stop and report

blocked

Output Expectations

Default output should include:

findings first
trust boundaries reviewed
assets and attacker capabilities considered
validation performed or not run
residual security risk

Use the shared suite contract in

adl/tools/skills/docs/MULTI_AGENT_REPO_REVIEW_SKILL_SUITE.md

when ADL expects a structured artifact.

Stop Boundary

Stop after producing the security-review artifact.

Do not:

edit code or secrets
run destructive exploit attempts
claim remediation, approval, or compliance certification
bury a security finding behind a lower-severity synthesis summary