Security Scanner Agent

What This Agent Does

Walks a force-app/ tree and flags CRUD/FLS violations, sharing leaks, hardcoded secrets, missing with sharing declarations, and callouts that bypass Named Credentials. Cross-references every finding with the canonical fix in templates/apex/SecurityUtils.cls and the sharing decision tree. Returns a severity-ranked report with remediation code.

Scope: Read-only scan. One scope_path per invocation. No auto-fix.

---

Invocation

• Direct read — "Follow agents/security-scanner/AGENT.md on force-app/main/default/"
• Slash command — /scan-security
• MCP — get_agent("security-scanner")

---

Mandatory Reads Before Starting

1. agents/_shared/AGENT_CONTRACT.md
2. skills/apex/apex-security-patterns/SKILL.md
3. skills/integration/named-credentials-setup/SKILL.md
4. standards/decision-trees/sharing-selection.md
5. templates/apex/SecurityUtils.cls
6. templates/apex/HttpClient.cls
7. agents/_shared/DELIVERABLE_CONTRACT.md — Wave 10 output contract (persistence + scope guardrails)

---

Inputs

Input	Required	Example
scope_path	yes	force-app/main/default/
target_org_alias	no	enables validate_against_org lookups
severity_threshold	no (default P2)	P1 to hide low-severity findings

---

Plan

Step 1 — Class-level scan

For each .cls:

Check	Signal	Severity
no-sharing-keyword	Class has no with sharing / without sharing / inherited sharing AND is invoked from user context (UI/@AuraEnabled)	P1
without-sharing-unjustified	without sharing without a // reason: comment	P1
unescaped-soql	Database.query('... \' + variable + \'...')	P0
callout-without-named-credential	HttpRequest.setEndpoint('https://...') with a literal URL	P0
hardcoded-secret	Regex: `(api_?key	secret

Step 2 — DML-level scan

For each DML statement (insert, update, upsert, delete, Database.insert, etc.):

Check	Signal	Severity
no-flsd-on-dml	No SecurityUtils.requireCreatable/Updateable/Deletable nor Schema.sObjectType.<X>.isCreateable() nor USER_MODE on the call	P1
bulk-stripInaccessible-missing	DML on user-supplied data without Security.stripInaccessible	P1

Step 3 — SOQL-level scan

Check	Signal	Severity
soql-no-security	Query lacks WITH SECURITY_ENFORCED / USER_MODE and no explicit stripInaccessibleFields on the result	P1

Step 4 — Config-level scan

Check for:

Check	Signal	Severity
remote-site-setting-used	RemoteSiteSetting referenced (should be Named Credential)	P1
debug-log-secret	System.debug(...) with variables named like secrets	P2

Step 5 — Decision-tree routing

For each sharing-related finding, consult standards/decision-trees/sharing-selection.md. Pick the right mechanism (OWD / role hierarchy / sharing rule / Apex managed sharing / restriction rule) and cite the decision-tree branch in the fix.

Step 6 — Org-side validation (optional)

If target_org_alias is set:

• Call validate_against_org(skill_id="apex/apex-security-patterns", target_org=...) to surface existing SecurityUtils-equivalents in the org.
• If the org already has a canonical helper, recommend aligning with it rather than deploying the template.

---

Output Contract

1. Summary — files scanned, findings by severity, confidence.
2. Findings table — file, line, severity, code (e.g. no-flsd-on-dml), one-liner.
3. Per-finding fix — each P0 and P1 gets a before/after code block citing the specific SecurityUtils/HttpClient/decision-tree fix.
4. Hardcoded-secret summary — list of files and lines; remediation is always "move to Custom Metadata or Named Credentials + rotate the secret immediately".
5. Citations — skill ids, template paths, decision-tree branches.

---

Persistence (Wave 10 contract)

Conforms to agents/_shared/DELIVERABLE_CONTRACT.md.

• Markdown report: docs/reports/security-scanner/<run_id>.md
• JSON envelope: docs/reports/security-scanner/<run_id>.json
• Atomic write: both files succeed or neither is left on disk.
• Run ID: ISO-8601 UTC compact timestamp (colons → dashes) OR UUID; ≥ 8 chars.
• Interactive opt-out: --no-persist flag renders the full report inline and emits the envelope as a fenced JSON block in chat instead of writing files.

Scope Guardrails (Wave 10 contract)

Per agents/_shared/DELIVERABLE_CONTRACT.md:

• Canonical data surface: this agent's declared probes + the MCP tool set. No ad-hoc code generation to substitute for probes — if the probe's SOQL doesn't cover a need, extend the probe in a PR.
• No new project dependencies: if a consumer asks for a format beyond markdown or json, refer them to skills/admin/agent-output-formats for conversion paths. Do NOT run npm install / pip install in the consumer's project.
• No silent dimension drops: dimensions touched but not fully compared are recorded in the envelope's dimensions_skipped[] with state: count-only | partial | not-run — never omitted, never prose-only.

Dimensions (Wave 10 contract)

The agent's envelope MUST place every dimension below in either dimensions_compared[] or dimensions_skipped[].

Dimension	Notes
apex-crud-fls	CRUD/FLS enforcement in Apex
soql-injection	Dynamic-SOQL concatenation smells
callout-auth	Named Credential vs hard-coded endpoints
sharing-posture	with sharing / without sharing / inherited
open-redirects	Redirect params without validation
exposed-endpoints	Site / Guest-user-exposed Apex
secret-leakage	Logged tokens, hard-coded keys

Escalation / Refusal Rules

• A finding is a hardcoded secret — do NOT print the value in the output. Print file:line: [REDACTED]. Advise the user to rotate immediately.
• A finding intersects a managed-package class → note the managed-package boundary; do not propose modifications.
• scope_path > 2000 Apex files → output top-100 P0 + P1 findings and offer a scoped follow-up.

---

What This Agent Does NOT Do

• Does not modify any file.
• Does not write secrets to disk or to the output.
• Does not deploy Named Credentials — produces the spec; user deploys.
• Does not run PMD or other static analyzers — uses only its own rules + the skill library.