CtrlK
BlogDocsLog inGet started
Tessl Logo

excessive-agency

Hunt LLM excessive agency (OWASP LLM06:2025) — agentic systems granted too many tools, too broad permissions per tool, or unsupervised authority to act on the user / business behalf, producing financial loss, data loss, or destructive operations from a single bad token.

68

Quality

82%

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

SKILL.md
Quality
Evals
Security

Quality

Content

90%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, concise offensive-security skill body with concrete commands, payloads, and a verification contract. The only meaningful gap is an inline validation/feedback checkpoint within the audit workflow itself.

Suggestions

Add an explicit validation step to the audit workflow (e.g. 'For each flagged tool, re-run the approval-logic grep and confirm the gate is genuinely absent before reporting') to create a validate→fix→retry loop.

Consider moving the CVSS table and PoC payloads into a reference file referenced one level deep, keeping SKILL.md as a tighter overview.

DimensionReasoningScore

Conciseness

Lean and efficient throughout — assumes Claude's familiarity with LLM security and OWASP, with no padding; every section (recognition signals, attack vectors, audit greps, PoCs, CVSS) earns its place.

5 / 5

Actionability

Provides copy-paste-ready grep commands, concrete PoC payloads with exact parameters, named tools, and a validate_finding contract with explicit success and negative patterns — fully executable guidance covering common cases.

5 / 5

Workflow Clarity

The audit workflow is clearly sequenced (enumerate inventory → destructive verbs → approval logic → creds → per-tool questions) and a verification contract exists in section 6, but the audit steps themselves lack an inline validate→fix→retry checkpoint, leaving a minor validation gap.

4 / 5

Progressive Disclosure

A single well-organized SKILL.md with eight clearly numbered sections and no nested references; structure is good and navigation is easy, though the CVSS table and PoC payloads could optionally be split into reference files for a fully one-level-deep layout.

4 / 5

Total

18

/

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 specific, well-targeted description with strong concrete language, but it lacks an explicit 'Use when...' trigger clause, relying on metadata for when-to-use guidance. Adding an inline trigger phrase would raise completeness.

Suggestions

Append an explicit 'Use when...' clause to the description, e.g. 'Use when auditing agentic LLM systems for excessive tooling, broad per-tool permissions, or missing approval gates on destructive operations.'

Add common trigger synonyms users say naturally, such as 'human-in-the-loop', 'approval gates', and 'tool scope', to broaden keyword coverage.

DimensionReasoningScore

Specificity

Names the domain (OWASP LLM06:2025) and lists multiple concrete failure modes and outcomes — 'too many tools', 'broad permissions per tool', 'unsupervised authority', 'financial loss, data loss, or destructive operations from a single bad token' — giving comprehensive coverage.

5 / 5

Completeness

Provides a strong 'what' but no explicit 'Use when...' trigger clause — the 'when' guidance lives only in the separate when_to_use metadata, so per the missing-trigger-clause cap completeness cannot exceed 3.

3 / 5

Trigger Term Quality

Includes natural terms like 'excessive agency', 'agentic tools', 'broad permissions', and 'unsupervised authority', but omits common variations such as 'human-in-the-loop', 'approval gates', or 'tool scope' that users would readily say.

4 / 5

Distinctiveness Conflict Risk

Targets a clear niche (OWASP LLM06 excessive agency) with specialized triggers that distinguish it from LLM01 prompt-injection or other AI-security skills, giving minimal conflict risk.

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.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

metadata_version

'metadata.version' is missing

Warning

Total

15

/

16

Passed

Repository
PurpleAILAB/Decepticon
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.