
Open-source security review skill for developers building with AI. Evidence-first reviews, fixes
Open-source security review skill for developers building with AI. Evidence-first reviews, fixes, and retests. Apache-2.0. By EL1Syum.
·
Evidence-first security review for developers building with AI coding agents.
KodeWaspada is an actively developed open-source blue-team security review agent/skill for OpenAI Codex and coding assistants that support compatible Agent Skills. It is designed to make code-security review less dependent on confident-sounding guesses and more dependent on traceable evidence.
Created and maintained by EL1Syum (@gustiakmal52). Free to use, modify, and distribute, including commercially, under Apache License 2.0 starting with 0.3.0.
The core workflow is:
issue → location → evidence → impact → fix → retest
KodeWaspada is an independent community project. It is not an official OpenAI or Daybreak product, certification, endorsement, or security guarantee.
A generic "act as a security expert" prompt can produce useful ideas, but it does not define how the agent should distinguish a real vulnerability from a plausible-looking false positive.
KodeWaspada gives the agent a repeatable decision process:
The only required component is skills/kodewaspada. Python and Semgrep are not required to use the skill.
From this repository:
mkdir -p /path/to/your-project/.agents/skills
cp -R -i skills/kodewaspada /path/to/your-project/.agents/skills/Then start a new coding-agent session and ask:
$kodewaspada Review this project before deployment.
Focus on authentication, cross-user data access, and dangerous input flows.
Show evidence, impact, and uncertainty. Do not modify code yet.The folder follows the Agent Skills structure used by current OpenAI tooling. Host-specific installation paths can differ, so review the skill before enabling it in your environment.
| Mode | Expected behavior |
|---|---|
| Threat mapping | Identify sensitive assets, actors, trust boundaries, and unknown areas |
| Code / diff review | Trace input → validation/authorization → sensitive operation → effect |
| Finding triage | Separate source evidence, local test evidence, unresolved hypotheses, and dismissed findings |
| Fix | Apply a small root-cause patch while preserving unrelated user changes |
| Retest | Run relevant regression checks and report actual results and remaining gaps |
Review guidance covers authentication/session handling, object/tenant/RLS authorization, injection, XSS, files/uploads, SSRF, secrets, dependencies, CI/configuration, AI-tool boundaries, and business logic.
That is review scope, not a claim that every category is automatically detected.
KodeWaspada intentionally separates confidence from impact.
| Evidence state | Meaning |
|---|---|
| Source-proven | The source-to-effect path and relevant control are visible in reviewed code |
| Locally tested | A local test or PoC was actually executed with a meaningful comparison control |
| Needs verification | A plausible path exists, but required helper/config/runtime/permission evidence is missing |
| Dismissed | A real control breaks the suspected path; it should not remain in the active finding count |
A source-level finding is not automatically a production exploit. A clean result is not a security certificate.
Suppose user A can request a note belonging to user B.
KodeWaspada should not stop at "the URL contains an ID." It should trace the request through middleware and the data query, inspect ownership checks, and distinguish code evidence from runtime evidence.
If the query already binds the object to the authenticated owner, the suspected cross-user access issue should be dismissed, not reported just because the route looked suspicious.
1. Review without editing
$kodewaspada Review this application folder.
Focus on login, cross-user data access, and database/HTML input.
Show evidence and limitations. Do not edit files.2. Fix a sufficiently supported finding
$kodewaspada Fix the cross-user data-access issue we just confirmed.
Preserve the public API and my unrelated changes.
Add owner vs. non-owner regression tests.3. Retest the patch
$kodewaspada Retest this patch.
Run the relevant local tests, compare allowed and denied access,
and report the actual results.A review request does not automatically authorize edits, deployment, publishing, opening PRs, or production testing.
KodeWaspada also includes an optional Python CLI that can run without AI. It currently provides three JavaScript/TypeScript taint checks for request-controlled input reaching:
It generates HTML, Markdown, and JSON reports.
python3 -m kodewaspada scan /path/to/project --out outputs/first-reviewThe scanner is a candidate finder, not the full security review. A zero-result scan does not mean the project is safe.
See docs/SCANNER.md for supported extensions, limits, privacy behavior, exit codes, and current engine constraints.
The repository includes synthetic evaluation fixtures for testing agent decisions, not just scanner syntax.
Current eval scenarios check whether the agent can:
See evals/README.md.
These fixtures are not a universal benchmark, an accuracy percentage, or proof that the skill has no false negatives. Material instruction changes should be reevaluated.
On one authorized private application checkout, the skill-led manual review found a file-boundary flaw and validated the behavior locally with a synthetic marker. No secret was accessed and no production behavior was claimed. Separately, a fresh evaluator correctly distinguished the equivalent vulnerable and safe functions in the public synthetic fixture.
The optional JS/TS scanner reported zero candidates within its supported frontend scope. That distinction is intentional: the finding came from manual agent review, not from pretending three JS/TS rules cover an entire application. One pilot demonstrates a useful workflow, not cross-repository accuracy. Application identity and exploit details remain private pending remediation.
Potential vulnerabilities in KodeWaspada itself should follow the private reporting process in SECURITY.md.
python3 -m unittest discover -s tests -v
python3 -m compileall -q kodewaspadaFull scanner coverage requires the documented Semgrep dependency. See docs/SCANNER.md and docs/QA.md.
External review, pull requests, and proposals are welcome, especially when they improve the evidence model rather than merely increasing the number of warnings. Contributions use Apache License 2.0, and contributors retain their copyright. See CONTRIBUTING.md.
Useful proposals include:
Read CONTRIBUTING.md before submitting changes.
For discussion, see the open feedback issue: real-world AppSec review cases.
If KodeWaspada improves your security-review workflow, consider starring the repository so other developers can find it.
Starting with 0.3.0, companies, consultancies, and paid products can use KodeWaspada under Apache License 2.0 without purchasing a license or requesting separate permission. Commercial redistribution is allowed under the same license obligations.
Paid onboarding, customization, or support can be agreed separately; none is required to use the project, and no support or SLA is included by default. See commercial use and earlier releases.
Starting with 0.3.0, KodeWaspada's original instructions, wrapper, rules, examples, and documentation are available under Apache License 2.0:
Copyright 2026 EL1SyumYou may use, modify, and redistribute the project, including in commercial products. When redistributing, include LICENSE, retain applicable attribution and NOTICE content, and identify modified files as required by Apache-2.0. The license does not grant trademark rights or imply endorsement. Earlier releases retain the license shipped with them; see the transition note.
Semgrep is a separate upstream dependency and is not bundled as the project's own engine.
Created and maintained by: EL1Syum (@gustiakmal52)
The coding assistant applies the skill to the requested scope: discover risky surfaces, trace input through real controls to an effect, classify the evidence, and report impact and coverage. Local validation depends on authorization and an available test environment. Fix and Retest follow the requested mode.
The four evidence labels describe what is supported; they are separate from severity. The optional JS/TS scanner supplies candidates for review through three rules: shell execution, eval / Function, and raw HTML output.
This shows one way a host coding assistant can distribute review work when it supports multiple agents. Roles, scheduling, and tool access come from the host; KodeWaspada provides the review guidance and evidence standard. A single coding assistant can use the same skill.
See the visual guide and source map for the text equivalents, requested Fix/Retest paths, and the implemented scanner pipeline. Panduan Bahasa Indonesia.
This repo does not document an install step we could read. Check the README on GitHub.
Paste this prompt into your agent. It reads this page and does the setup for you.
Read https://aiagentslisting.com/mcp/kodewaspada to learn what the "KodeWaspada" MCP server does and how to install it. Add it to my coding agent's MCP configuration as documented on that page, then confirm the server connects and list the tools it exposes.Agents can also browse this directory over MCP at https://aiagentslisting.com/api/mcp
This server doesn't publish a machine-readable tool manifest, and it doesn't expose a public endpoint we could ask. Connect the server locally to see what it exposes.
We could not run this server: it ships no npx package for our sandbox and no public endpoint. The four checks that need a live server are left out, and the grade tops out at B until one exists.
Maintain this server? Claiming proves you own the listing and earns the last 4 points.
Claim this listingPaste this line near the top of your repository's README. The badge always renders the current score.
[](https://aiagentslisting.com/mcp/kodewaspada)Nothing comparable is listed yet.
v0.3.0KodeWaspada 0.3.0 — Open SourceRelease notesv0.2.1KodeWaspada 0.2.1Release notesConnect to the AI Agents Listing MCP server at https://aiagentslisting.com/api/mcp using the HTTP transport and complete the sign-in when it asks. Then call the submit_review tool for the listing "kodewaspada" with a rating from 1 to 5 and a short note on what worked and what did not.No reviews yet
Be the first to share your experience with this tool.
One email a week. New agents, MCP servers and skills, and what is actually getting traction.