hashgraph-online/awesome-codex-plugins / plugins/boshu2/agentops/skills/security
security
FlaggedReview code for security problems; scan for vulnerabilities, secrets, dependency and prompt risks. Use when: asked whether code is safe to ship, even one small handler.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 10.1 KB |
| references/agentops-redteam-pack.json | config | 6.4 KB |
| references/attacker-path-evidence.md | markdown | 4.7 KB |
| references/owasp-checklist.md | markdown | 6.7 KB |
| references/policy-example.json | config | 525 B |
| references/security-suite-runbook.md | markdown | 8.4 KB |
| scripts/prompt_redteam.pyexecutable | script | 11.3 KB |
| scripts/security_suite.pyexecutable | script | 32.7 KB |
| scripts/validate.shexecutable | script | 5.4 KB |
references/owasp-checklist.md
View on GitHub# Security Class Checklist
> Code-level review classes drawn from the OWASP Top 10 plus SSRF, deserialization,
> command injection, cryptography and prompt injection. Load it during a `/security`
> review pass to walk each class and record a per-class result. It is not exhaustive:
> add any class the surface needs (a race, business logic) to the ledger rather than
> calling it clean. It ranks findings by severity; it does not gate merges or
> releases — those are caller decisions.
## Checklist
### 1. Secrets Management
- [ ] No hardcoded API keys, passwords, or tokens in source
- [ ] All secrets loaded from environment variables or secret stores
- [ ] `.env` files in `.gitignore`
- [ ] No secrets in log output, error messages, or terminal echo while a person types one in
- [ ] CI/CD secrets use platform-native secret management, and each job receives only the secrets its workflow references
- [ ] When one operation issues several credentials, routine calls use the least-privileged one and the elevated one is kept only for the operations that need it
**Detection:**
```bash
grep -rn 'password\s*=\s*"[^"]\+"\|api_key\s*=\s*"[^"]\+"\|secret\s*=\s*"[^"]\+"\|token\s*=\s*"[^"]\+' --include='*.go' --include='*.py' --include='*.ts' --include='*.js' . | grep -v _test | grep -v test_ | grep -v vendor/ | cut -d: -f1-2 # file:line only; never print the matched value
```
### 2. Input Validation
- [ ] All user input validated with schema (Zod, JSON Schema, struct tags)
- [ ] Input length limits enforced
- [ ] Content-type validation on file uploads
- [ ] No `eval()`, `exec()`, or dynamic code execution with user input
- [ ] Path traversal prevention: resolve a user-supplied path (following symlinks, or open it through a root-bound API such as Go's `os.Root`), then confirm by path components that it stays inside the allowed directory. A string-prefix compare or a ban on a `..` substring is not that check: `/srv/base-evil` starts with `/srv/base`, and `foo/../bar` and `a..b` stay inside
- [ ] Secret files and scratch in a shared temp directory are owner-only from the moment they exist (restrictive umask before creation, not chmod after the write); scratch takes a fresh `mktemp` path, and code that reuses a fixed, guessable name refuses it when it is a symlink or another user's entry at the open itself (exclusive no-follow create: `O_EXCL|O_NOFOLLOW`, or `set -C` in shell), not by a test before the write, since any local user can create or redirect that name first
### 3 to 10. Standard web classes
Give each its own ledger result.
- 3. SQL injection (any query text built from input, raw queries included)
- 4. XSS (output encoding, `innerHTML`-style sinks, CSP)
- 5. CSRF (state-changing requests, cookie `SameSite`, origin checks)
- 6. Authentication (token storage, session expiry, password hashing, auth rate limits)
- 7. Authorization (server-side check on every endpoint and object, admin paths)
- 8. Rate limiting (public, auth and payment endpoints)
- 9. Sensitive data exposure (logs, error bodies, API fields, transport, at rest)
- 10. Dependencies (known-vulnerable versions, pinning, committed lock files)
### 11. CI and Automation Definitions
- [ ] Triggers that run contributor-controlled code (forked pull requests, comment commands) get no write token or secrets
- [ ] Attacker-controlled fields (PR title, branch name, issue body) never expand directly into shell or script steps. The same holds wherever a script or agent instruction writes a command around text it did not author (a branch name from PR data, a model id or effort level from config): admit the value only if it matches a narrow character allowlist, or keep it out of the command text by passing it through a variable, file or stdin, and refuse a failing value rather than quoting it harder. `$(...)` still runs inside double quotes, and git accepts it in branch names
- [ ] Third-party actions, images and scripts are pinned to an immutable commit or digest, not a moving tag
- [ ] Each job's token permissions are the minimum it needs; default write access is off
- [ ] Every job has an execution time limit
### 12. Server-Side Request Forgery (SSRF)
- [ ] A server-side fetch of a caller-supplied URL or host is limited to an allowlist; redirects and resolved addresses are rechecked, and loopback, private and cloud-metadata addresses are refused
### 13. Deserialization
- [ ] Untrusted bytes are never loaded with a format that can build arbitrary objects (`pickle`, `yaml.load` without a safe loader, Java native serialization, `Marshal`); use a data-only format and validate it
### 14. OS Command Injection
- [ ] Application code never builds a shell command from untrusted input: pass an argument list (no `shell=True`, `os.system`, or `sh -c` around interpolated text) and allowlist any value that becomes an argument; class 11 covers scripts and agent instructions
### 15. Cryptography
- [ ] Standard libraries and current algorithms only (no ECB mode, no MD5 or SHA-1 for integrity, no home-made scheme); keys and IVs are never hardcoded or reused; tokens come from a CSPRNG (`secrets`, `crypto/rand`), not `random`; TLS certificate verification is never disabled
### 16. Prompt Injection and Model Output
- [ ] Untrusted text that reaches a model (retrieved documents, issues, web pages, tool output) cannot widen what it may do: tool access, secrets and write actions are bounded by the caller's authority, not by the prompt
- [ ] Model output is untrusted input before it reaches a shell, query, file path, HTML or another tool
## Severity Classification
Severity ranks findings so a reviewer can order them; it carries no merge, release,
or remediation-timing authority. Whether and when to fix, and whether to block any
delivery, are caller decisions this checklist does not make.
| Finding | Severity |
|---------|----------|
| Hardcoded secret in source | CRITICAL |
| SQL injection possible | CRITICAL |
| OS command injection or unsafe deserialization of untrusted input | CRITICAL |
| SSRF reaching internal or metadata addresses | HIGH |
| Missing input validation on public endpoint | HIGH |
| Dependency with known CVE (CVSS > 7) | HIGH |
| Missing rate limiting | MEDIUM |
| Missing CSP headers | MEDIUM |
| Debug logging in production code | LOW |
## Integration
### With /security (suite primitives)
The offline redteam scan (`prompt_redteam.py scan`) checks repo-owned prompt and control surfaces against the attack pack. It does not review application code, so every item in this checklist still needs a code-level result: finding, clean, or not assessed.
### With CI
```bash
# Minimum: secrets + dependencies
grep -rn 'password\|secret\|api_key' --include='*.go' --include='*.py' . | grep -v test | cut -d: -f1-2 # file:line only
govulncheck ./... # or npm audit / pip audit
```
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first. | Stars: 194k | Updated | Claude CodeCodex | Verified | Install |