CtrlK
BlogDocsLog inGet started
Tessl Logo

aatmf-t10-confidentiality-breach

AATMF T10 — Integrity & Confidentiality Breach. System prompt extraction, training-data extraction, model-weight leakage, private-key recovery.

56

Quality

63%

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/plugins/llm-redteam/t10-confidentiality-breach/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

68%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.

The body is a lean, well-structured attack-technique catalog with highly actionable copy-paste prompts and a concrete probe configuration. Its main weakness is the absence of an explicit sequenced workflow with validation checkpoints tying the probe, detection, and severity sections together into a feedback loop.

Suggestions

Add a short ordered workflow (run probes → check detection signals → rate severity → escalate/descop) with an explicit validation checkpoint before reporting a finding.

Move the long list of T10.001 prompt variations into a references file and keep a concise representative subset inline, enabling one-level-deep progressive disclosure.

Tighten the prose asides in T10.005 ("Theoretical…", "Rarely practical…") to bullet form to recover token efficiency.

DimensionReasoningScore

Conciseness

The body is lean — compact bullet lists of copy-paste prompts and a YAML config assuming Claude knows the domain — with only minor prose asides ("Theoretical…", "Rarely practical due to query economics", "The big one.") that could be trimmed, fitting anchor 4 rather than the every-token-earns-its-place anchor 5.

4 / 5

Actionability

Provides concrete copy-paste attack prompts (ten numbered variations) and a concrete YAML probe config covering the common cases, but T10.005 is largely descriptive and there is no automation script, leaving minor gaps that fit anchor 4 over the fully-executable anchor 5.

4 / 5

Workflow Clarity

This is a reference catalog, not a sequenced workflow; only a light "try in order of subtlety" sequence exists, and the batch probe operation has no integrated validation checkpoints (detection signals live in a separate section), so the missing-validation cap of 3 applies.

3 / 5

Progressive Disclosure

No bundle files exist and the single self-contained file is well-organized with clear section headers; at ~115 lines it exceeds the under-50-line exception and has no one-level-deep references, so it earns the good-structure anchor 4 rather than the reference-split anchor 5.

4 / 5

Total

15

/

20

Passed

Description

58%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.

The description clearly states what the skill covers via four concrete extraction actions, but omits any explicit "Use when…" trigger guidance and leads with framework codename jargon, capping completeness and trigger-term quality. It is distinct within a niche but has minor overlap with adjacent AI-security tactics.

Suggestions

Add an explicit "Use when…" clause naming natural user triggers (e.g., testing whether a model leaks its system prompt, training data, or embedded secrets).

Drop or de-emphasize the "AATMF T10" codename from the description so it leads with plain-language capability terms a user would actually say.

Include common synonyms/file-shape terms (e.g., "API keys", "secrets", "PII") to broaden natural trigger coverage.

DimensionReasoningScore

Specificity

Lists four concrete actions ("System prompt extraction, training-data extraction, model-weight leakage, private-key recovery") but they are outcome-categories rather than granular operations, so it sits just below the comprehensive anchor 5.

4 / 5

Completeness

The "what" is clear (the four extraction actions) but there is no "when"/"Use when…" clause, so per the cap rule completeness cannot exceed 3 — matching the clear-what-missing-when anchor exactly.

3 / 5

Trigger Term Quality

Contains relevant keywords (system prompt, training data, private key) but opens with the codename "AATMF T10" jargon and lacks natural "Use when…" phrasing or synonyms, matching the some-keywords-but-missing-variations anchor.

3 / 5

Distinctiveness Conflict Risk

It carves a specific confidentiality-breach niche but overlaps with related AI-security tactics (T1/T7/T11 cross-referenced in the body), fitting the mostly-distinct-minor-overlap anchor rather than the minimal-conflict anchor 5.

4 / 5

Total

14

/

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.