CtrlK
BlogDocsLog inGet started
Tessl Logo

aatmf-t04-memory-manipulation

AATMF T4 — Multi-Turn & Memory Manipulation. Persistent memory injection, conversation-state poisoning, cross-session contamination, ghost-context leak.

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/t04-memory-manipulation/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 content is a compact, well-structured red-teaming technique catalog with concrete example payloads and a probe config, but its test workflow lacks an explicit sequenced validation loop and the custom multi-turn harness is not provided as executable code.

Suggestions

Add a short numbered test workflow with explicit validation checkpoints (e.g., craft probe → run multi-turn → check detection signals → confirm severity → retry on miss).

Provide an executable python_provider multi-turn harness skeleton instead of only noting that one is needed.

Tighten or relocate the two-line intro so the body assumes Claude's competence throughout.

DimensionReasoningScore

Conciseness

The body is lean and uses shorthand ('RAG-w/-history', '→', 'doc w/'), with only minor over-explanation in the two-line intro that frames concepts Claude already knows.

4 / 5

Actionability

Concrete example payloads, a promptfoo YAML probe config, detection signals, and defender mitigations are provided; the gap is that the custom multi-turn test harness is described but not supplied as executable code.

4 / 5

Workflow Clarity

The technique catalog implies a test sequence (probe → detect → severity), but checkpoints are implicit via detection signals and there is no explicit validate→fix→retry feedback loop.

3 / 5

Progressive Disclosure

A well-organized single file with clear section headers (Techniques, Probe pattern, Detection signals, Severity, Defender, Cross-references) and no external references needed; at ~90 lines it slightly exceeds the 50-line simple-skill exception.

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 names a specific red-teaming niche and lists concrete techniques, but it lacks an explicit 'Use when...' trigger clause and relies on jargon-heavy keywords rather than natural user phrasing.

Suggestions

Add an explicit 'Use when...' clause stating when Claude should invoke this skill (e.g., 'Use when testing persistent memory, cross-session, or multi-turn prompt-injection attacks').

Soften jargon with natural synonyms users actually say, such as 'prompt injection', 'RAG poisoning', or 'memory poisoning'.

Reinforce distinctiveness from T1/T12 by foregrounding the multi-turn/stateful angle in the description itself.

DimensionReasoningScore

Specificity

Lists several concrete techniques ('Persistent memory injection, conversation-state poisoning, cross-session contamination, ghost-context leak'), naming the domain plus multiple specific actions with only minor coverage gaps.

4 / 5

Completeness

The 'what' is clear (memory manipulation techniques) but there is no 'Use when...' trigger clause, so per the rubric guideline completeness is capped at 3.

3 / 5

Trigger Term Quality

Relevant domain keywords are present ('memory injection', 'cross-session', 'conversation-state poisoning') but they lean technical and miss natural synonyms like 'prompt injection' or 'RAG poisoning' that a user would say.

3 / 5

Distinctiveness Conflict Risk

The 'AATMF T4 — Multi-Turn & Memory Manipulation' niche is mostly distinct, but it overlaps with related T1 prompt-injection and T12 RAG-poisoning tactics noted in the cross-references.

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.