feat(web-security): expand SQLi-to-RCE, math-rendering, and XS-Leak coverage

[GangGreenTemperTatum]

Summary

Evaluated recent public web-security research and integrated the reusable techniques that filled genuine gaps in the capability playbook set. Markdown-only changes (new/updated SKILL.md files) plus a version bump — no tooling, MCP, or runtime changes.

Three additions, each vetted against the existing 79 skills to avoid duplication:

1. sqli-to-rce-escalation (new skill)

Escalates a confirmed SQL injection into OS command execution or a planted webshell. blind-sqli-extraction already covers reading data through an oracle, but nothing covered the escalation-to-RCE path. DBMS-complete:

• Privilege preflight per engine (MSSQL sysadmin, PostgreSQL rolsuper / pg_execute_server_program, MySQL FILE + secure_file_priv semantics, Oracle DBA, embedded-Java owners) — the escalation is gated on the DB role, not the injection.
• OS-reach primitives: xp_cmdshell, OLE Automation, COPY TO PROGRAM, pg_cron non-stacked ORDER BY reach, INTO OUTFILE/DUMPFILE UDF, Derby SYSCS_EXPORT_QUERY → JSP webshell, H2 CREATE ALIAS.
• Execution-context handling: MSSQL Msg 574 user-transaction COMMIT breakout, second-order write/read split, terminator + URL-encoding traps.

2. math-rendering-macro-injection (new skill)

Abuse user-controlled TeX macros in MathJax/KaTeX renderers (GitHub-class content-integrity bug). Covers newcommand-package operator shadowing (priority -1 CommandMap, no base lookup), KaTeX \def/\href trust and \html* sinks, macro-expansion DoS (maxMacros/maxExpand), impact framing, and remediation configs. No prior skill touched math-rendering surfaces.

3. browser-side-channel (extended)

Added the ORB status-code XS-Leak revived via a service-worker fetch proxy that rewrites Sec-Fetch-Dest to empty, turning ORB network errors into blank 200s so the <script> onload/onerror status-code oracle works again in modern Chrome/Firefox. Complements the existing connection-pool / ETag / timing techniques.

Evaluated but intentionally not added:

• postMessage targetOrigin IP-normalization bypass — already covered in dom-vulnerability-detection.
• npx-confusion / bin-mismatch supply-chain RCE — out of scope for web-app pentest; belongs in the sast / supply-chain capability, not web-security.

Validation

• dreadnode capability validate capabilities/web-security → passes (81 skills; the two caido-cli/burp warns are pre-existing external-CLI checks, confirmed identical on clean main).
• Core technical claims spot-checked against authoritative docs (KaTeX trust:false forbids \href; MathJax packages: {'[-]': [...]} removal syntax).
• Peer-review pass on the SQLi skill produced fixes (corrected secure_file_priv NULL = export-disabled, added pg_execute_server_program, Oracle external-job and Derby CSV-quoting caveats, References + Checkpoints).

Bumps web-security 1.8.0 → 1.9.0.