feat(skills): expand llm_prompt_injection with Model DoS and Excessive AgencyFeat/llm dos excessive agency

[MujumdarSahil]

Expand llm_prompt_injection.md: Model DoS + Excessive Agency

Addresses #691 (AI/LLM application testing skills).

Prompt injection, jailbreaks, and unsafe output handling were already
covered by this skill (added in #616). This PR expands the same file
to cover two additional LLM runtime vulnerability classes that share
the confused-deputy testing methodology already established here:

• Model Denial of Service: context/token exhaustion, recursive tool
  call loops, chat-history state bloating
• Excessive Agency: missing authorization gates on destructive tool
  actions, privilege mismatch, confused-deputy tool execution

Both additions follow the file's existing structure (Attack Surface →
Key Vulnerabilities → Testing Methodology → Validation → Impact) rather
than introducing a new format, and include concrete testing probes
consistent with the rest of the skill.

Scope note: this does not address #678 (cross-modal/image-based
injection) — that requires genuinely different coverage (payloads
embedded in image pixels, not text metadata) and is left for a
separate PR.

Model theft and training-data poisoning (also part of OWASP LLM Top 10)
are intentionally excluded — they aren't dynamically testable via the
web-facing attack surface Strix operates on, so they don't belong in a
DAST-oriented skill file.

[MujumdarSahil]

@greptile-apps review

[greptile-apps]

Greptile Summary

This PR expands the LLM prompt injection skill with additional runtime security guidance.

• Adds Model DoS coverage for context exhaustion, loops, and state bloating.
• Adds Excessive Agency coverage for destructive tools and authorization gaps.
• Extends testing, validation, and impact sections for the new cases.

Confidence Score: 5/5

This looks safe to merge.

• No blocking issues found in the changed code.

Important Files Changed

Filename	Overview
strix/skills/vulnerabilities/llm_prompt_injection.md	Adds additive guidance for Model DoS and Excessive Agency while preserving the existing skill structure and frontmatter shape.

_{Reviews (1): Last reviewed commit: "feat(skills): expand llm_prompt_injectio..." | Re-trigger Greptile}