Skip to content

Kartik-Doye/Ravengard-AI-Recruiter / agent/skills/harden

harden

Verified

Threat-model, review, or reduce exploitable risk in software you are authorized to assess, including application code, APIs, identities, data flows, dependencies, CI/CD, configuration, cloud/IaC, and agentic capabilities. Use for security-sensitive changes, feature threat models, security audits, or evidence-backed remediation and retesting.

FileKindSize
SKILL.mdskill md18.0 KB
references/attack-surface.mdmarkdown5.1 KB
references/research.mdmarkdown6.0 KB

references/attack-surface.md

View on GitHub
# Attack-surface checklist: OWASP Top 10:2025 baseline, ASVS-depth where needed

Audit order follows the threat model, not this list's order. Each item:
what to check, and how it's commonly broken.

## A01: Broken access control

- Object-level checks per request (IDOR: change ids in URLs/bodies
  against your own accounts: does 404/403 actually happen?).
- Function-level: admin actions, bulk endpoints, internal-only routes
  actually gated. Deny-by-default, not allow-by-omission.
- CORS: no reflected arbitrary origins with credentials.
- DB permissions/RLS: app's DB user scoped to what it needs; row-level
  rules where multi-tenancy exists.
- Webhook/admin callbacks: signature or secret verified before action.

## A02: Cryptographic failures

- Passwords: argon2id/bcrypt, never MD5/SHA/plain, never reversible.
- Secrets: no hardcoded keys/tokens (grep history too: `git log -p`,
  tracked `.env`); rotation plan exists; keys not in client bundles.
- TLS everywhere internal where sensitive; no self-signed-in-prod
  surprises; secure cookie flags.
- Sensitive data minimized: don't store what you don't need (the cheapest
  protection for any leak).

## A03: Injection (Including OWASP GenAI)

- SQL/NoSQL built from user input (string concatenation, raw query
  fragments, unparameterized `IN (...)` lists).
- Command execution: user input reaching `exec`/shell/eval: including
  via file names and "safe" helpers.
- Template/HTML injection; unsafe deserialization of user-controlled
  blobs (pickle, untrusted YAML, PHP object injection).
- XSS: user content rendered without sanitization, `dangerouslySetInnerHTML`,
  `v-html`; DOM XSS via URL fragments into sinks.
- **AI/LLM Injection (OWASP GenAI):** **Indirect Prompt Injection (IPI)**: external data (PDFs, emails, DB records) processed by the agent must be treated as untrusted. Prompt injection is the new RCE; tool calls generated by models must be strictly validated.

## A04: Insecure design & Excessive Agency

- **Excessive Agency (OWASP GenAI):** Granting an AI agent broad, standing permissions (e.g. unmonitored write access to production DBs, emails) without human-in-the-loop approval gates or reversibility.
- Missing rate limits on auth, password reset, expensive endpoints.
- Business-logic abuse: negative quantities, self-referrals, double-submit,
  race conditions on limited inventory (try them: safe adversarial
  verification).
- Trust placed in client-side validation alone; price/quantity/role
  recalculated server-side.
- File uploads: type/size/content validated server-side, stored outside
  webroot, never executed, names sanitized (path traversal: `../` in
  filenames).

## A05: Security misconfiguration

- Debug/dev modes off in prod; verbose stack traces to clients off.
- Security headers: CSP (no unsafe-inline), HSTS, X-Content-Type-Options,
  frame-ancestors; cookies: Secure/HttpOnly/SameSite.
- Default credentials, sample apps, directory listing, exposed `.git`,
  `/.env`, admin panels with default passwords (check: it happens).
- Dependencies: audit tools run, deprecated packages flagged, lockfile
  committed; unknown transitive additions questioned (supply chain).

## A06: Authentication failures

- Session handling: rotation on login, invalidation on logout, expiry on
  idle + absolute; fixations tested.
- Credential recovery flows: reset tokens single-use, short-lived,
  unguessable; no user enumeration via differing responses/timing.
- OAuth/OIDC: state (CSRF), PKCE for public clients, iss/aud/nonce
  validated, redirect URIs exact.
- MFA where the threat model says so; login endpoints rate-limited with
  lockout/backoff that doesn't enable DoS on users.

## A07: Integrity failures

- Updates/CI: pinned action versions, minimal token scopes, no
  curl-pipe-bash in docs/CI, artifact provenance where the ecosystem
  supports it.
- Webhook payloads verified by signature before parsing; replay windows
  enforced.
- Unsigned auto-update paths, unpinned dependencies in deploy scripts.

## A08: Logging & monitoring failures (defenders' side)

- Auth events, permission failures, and admin actions logged with
  context: you can't respond to what you never recorded.
- Logs contain no secrets/PII payloads; log injection (newlines from
  input) handled.
- Alerts exist for the failure patterns that matter (auth-failure spikes,
  5xx rates): listed as recommendations if absent.

## SSRF & outbound (where the app fetches URLs)

- User-supplied URLs fetched? Validate scheme/host, block internal
  ranges (169.254.x, metadata endpoints, localhost), no redirect
  following into internal space, response size/time caps.

## Severity calibration

| Sev | Meaning |
| --- | --- |
| Critical | Exploitable remotely without auth → data breach, RCE, or money loss |
| High | Exploitable with modest conditions (an account, a crafted link) or serious impact |
| Medium | Real weakness, limited reach/impact, or strong mitigations present |
| Low | Hardening opportunity; exploitable only in contrived conditions |

Confidence is separate and always stated: `confirmed` (reproduced) /
`probable` (clear code path, not executed) / `speculative` (pattern
smell). A confirmed medium beats a speculative critical in triage order.
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 196kUpdated Claude CodeCodexVerifiedInstall