CtrlK
BlogDocsLog inGet started
Tessl Logo
SkillPublic

unbounded-consumption

Hunt LLM unbounded consumption (OWASP LLM10:2025) — denial-of-wallet and denial-of-service against LLM endpoints via unrestricted prompt size, runaway tool loops, expensive model selection, and unauthenticated fan-out.

Quality

80%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Critical

Do not install without reviewing

Fix and improve this skill with Tessl

tessl review fix ./packages/decepticon/decepticon/skills/standard/analyst/unbounded-consumption/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

78%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

A highly actionable, well-sequenced pentest playbook with copy-paste-ready PoCs, audit commands, and a validation contract. Its weaknesses are structural: everything lives in one monolithic SKILL.md with no reference files, and a small amount of redundant threat-model exposition at the top.

Suggestions

Split the PoC payload library (§5) and the CVSS table (§7) into one-level-deep reference files (e.g., references/payloads.md, references/cvss.md) with clear 'See X' links from SKILL.md, keeping the overview lean.

Add an explicit feedback loop to the audit workflow: what to do when a validate_finding success pattern fails to reproduce (e.g., re-check tier config, retry with a fresh account, downgrade severity), turning §6 from a contract into a checkpoint.

Trim or tighten the opening paragraph to just the three-impact escalation framing, dropping generalities about token pricing that Claude already knows.

DimensionReasoningScore

Conciseness

Nearly every section is signal — recognition signals, grep one-liners, payload tables — but the opening threat-model paragraph ('LLM inference is metered in dollars-per-token at the provider...') explains concepts Claude already knows, so it is efficient with a minor instance of over-explanation that could be trimmed, the 4 anchor rather than 5.

4 / 5

Actionability

Fully executable throughout: copy-paste-ready curl PoCs with env-var placeholders, concrete grep -rE audit commands, a ready-to-submit agentic-loop payload, a precomputed CVSS vector table, and a validate_finding contract with explicit success/negative patterns — matching the 5 anchor's 'copy-paste ready' standard.

5 / 5

Workflow Clarity

Numbered sections 1-8 form a coherent recognize → audit → exploit → validate → score → chain pipeline, with a validation checkpoint present (the validate_finding contract). It falls short of 5 because no explicit feedback loop (validate → fix → retry) is defined for the audit workflow itself, and the per-endpoint questions in §3 are checkpoints without stated follow-up on failure.

4 / 5

Progressive Disclosure

The skill is a ~160-line monolithic body with no bundle files at all: no references/, scripts/, or assets/ exist. Sections are well-organized, but content that clearly belongs in one-level-deep reference files (the PoC payload library §5, the CVSS table §7) is fully inlined — matching the 3 anchor 'content that should be separate is inline' rather than 2, since structure is present and clear.

3 / 5

Total

16

/

20

Passed

Description

75%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

A strong, specific description that names the vulnerability class and enumerates four concrete attack surfaces with a distinct niche identity. Its main weakness is the missing 'when to use' trigger clause inside the description itself (it lives only in metadata), and a few natural trigger-term variations are absent.

Suggestions

Append an explicit trigger clause to the description itself, e.g., 'Use when auditing LLM endpoints for cost/DoS abuse, token-budget bypasses, or runaway agentic loops' — the when_to_use metadata is not surfaced as the description and completeness is capped without it.

Add natural user phrasings to the trigger terms: 'cost attack', 'token budget', 'bill drain', 'rate-limit bypass', which a requester would plausibly say rather than the OWASP formal name.

DimensionReasoningScore

Specificity

The description lists multiple concrete attack vectors — 'unrestricted prompt size, runaway tool loops, expensive model selection, and unauthenticated fan-out' — under a named vulnerability class (OWASP LLM10:2025), giving comprehensive coverage with no gaps, matching the 5 anchor.

5 / 5

Completeness

The 'what' is clear and comprehensive (hunt unbounded consumption via four specific vectors), but the description itself contains no 'Use when...' clause or equivalent trigger phrase; the 'when' guidance exists only in the separate when_to_use metadata field, so per the judging guidelines completeness is capped at 3.

3 / 5

Trigger Term Quality

Good natural keyword coverage ('LLM', 'denial-of-wallet', 'denial-of-service', 'OWASP LLM10', 'prompt size', 'tool loop'), but common phrasings a user might actually say are missing (e.g., 'cost attack', 'token budget', 'bill drain', 'rate limit') — good coverage with a few natural terms missing, the 4 anchor.

4 / 5

Distinctiveness Conflict Risk

'LLM unbounded consumption (OWASP LLM10:2025)' carves a clear niche, and the generic-sounding DoS terms are qualified 'against LLM endpoints', keeping conflict risk with sibling security skills minimal — the 5 anchor.

5 / 5

Total

17

/

20

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository

PurpleAILAB/Decepticon

Path

packages/decepticon/decepticon/skills/standard/analyst/unbounded-consumption/SKILL.md

Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.