Eresus remediator

SAST Remediation

Purpose

Turn confirmed or likely security findings into safe, minimal, production-ready code changes. Prefer root-cause fixes over cosmetic mitigations, preserve legitimate behavior, and reduce the chance of regressions or bypasses.

Inputs

This skill works best when at least one of the following is available:

• a finding from eresus-sast-scanner
• a security review comment or PR thread
• a vulnerable file/function/endpoint
• a proof-of-concept payload or attack path

If eresus-sast-scanner is installed, load the matching vulnerability knowledge file(s) from its references/ directory before patching.

---

Workflow

Step 1: Reconstruct the Vulnerability Path

Before editing code, confirm:

• the attacker-controlled source
• the transformation path
• the vulnerable sink
• the intended business behavior that must still work after the fix

Do not patch from the label alone. Patch the actual code path.

Step 2: Choose the Right Fix Layer

Prefer the highest-leverage fix that closes the whole class of issue:

Problem Type	Preferred Fix
SQL / NoSQL injection	Parameterization, safe query builders, strict allowlists
XSS	Context-specific output encoding or safe templating APIs
SSRF	Destination allowlists, URL parser validation, egress restrictions
Path traversal	Canonicalization plus base-directory enforcement
Auth / IDOR	Server-side authorization at object/action boundary
CSRF	Framework-native CSRF defense, same-site cookies, token checks
File upload	Type/content validation, non-webroot storage, random filenames
Serialization / deserialization	Explicit DTOs, allowlisted types, safe parser settings, integrity checks
Weak crypto	Modern primitives, strong secrets, secure randomness
Race conditions	Atomic operations, locking, compare-and-set, transaction boundaries

Fix at the sink when possible, but move the fix earlier if many sinks share the same unsafe input path.

Step 3: Implement a Minimal, Safe Patch

While patching:

• preserve public API behavior unless the insecure behavior itself must change
• prefer framework-native security features over custom regex or denylist logic
• keep validation rules explicit and auditable
• update adjacent code when partial fixes would leave equivalent bypasses
• add concise comments only when the security reason is otherwise non-obvious

Step 4: Add Verification

Add or update tests when the repo supports them. Cover both:

• the blocked attack path
• the expected legitimate flow

If tests are not practical, still perform a manual reasoning pass that explains why the taint path is now broken.

Step 5: Re-Review the Fix

Run a short post-fix check:

• Can attacker-controlled data still reach the same sink?
• Did the patch move the problem instead of removing it?
• Does sanitization match the exact output context?
• Did the change accidentally create auth, compatibility, or logging regressions?

Step 6: Report the Remediation

Summarize:

• what was vulnerable
• what changed
• why the new control is effective
• what residual risk or rollout note remains

---

Remediation Guardrails

• Do not silence findings with comments, feature flags, or dead-code moves unless the execution path is truly removed.
• Do not replace parameterization with string escaping for query sinks.
• Do not trust client-side validation as the primary fix for a server-side issue.
• Do not add broad denylists when a precise allowlist or safer API exists.
• Do not "fix" IDOR by hiding identifiers in the UI; enforce authorization on the server.
• Do not fix SSRF with substring checks alone; parse and validate the normalized destination.
• Do not keep weak password hashing for backward compatibility without a migration or rehash plan.
• Do not claim a fix is complete if equivalent sibling endpoints remain exploitable.

---

Output Format

When reporting the remediation, use:

# Security Remediation Summary

## Finding
<short description of the original issue>

## Root Cause
<why tainted or untrusted data reached a dangerous operation>

## Fix
<what code changed and which control now blocks exploitation>

## Validation
<tests added, reasoning performed, or manual verification steps>

## Residual Risk
<migration note, rollout concern, or "none identified">