Back to skill

Security audit

meta-super-agent-bootstrap

Security checks for vulnerabilities and agentic risk

Overview

The skill is not overtly malicious, but it asks agents to persist and reuse preferences and even write learned behavior back into the skill without clear limits or consent.

Review this skill carefully before installing. Its main risk is persistent agent memory: it can record notes and preferences, reuse them later, and the instructions even contemplate changing SKILL.md based on accumulated experience. Install only if you are comfortable auditing and controlling that memory file, and avoid storing sensitive preferences or task details in it.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T02 · Agent Memory Poisoning

Warning
Location
scripts/learner.py:9
Finding

Unvalidated Persistent State Can Poison Future Agent Behavior

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:54-78; scripts/learner.py:9-17
Vulnerability Type: T02: Agent Memory Poisoning
Risk Level: Medium

The skill instructs the agent to retain free-form errors, usage notes, and user preferences in learned_patterns.json, apply important preferences during later invocations, and write accumulated experience back into SKILL.md. The implementation persists notes without validation or separation between passive data and behavioral instructions.

Relevant instructions from SKILL.md:54-78, translated into English:

text
Each use automatically reviews and accumulates experience, gradually improving
output quality and execution efficiency without manual maintenance.

learned_patterns.json records operation counts, capability usage frequency,
error patterns, user preferences, and improvement suggestions.

Important user preferences → write them into learned_patterns.json and directly
adopt them during the next invocation.

Errors accumulated at least three times → proactively add validation or fallback
steps and write the experience back into this SKILL.md.

Complete persistence logic from scripts/learner.py:9-17:

python
def record(op_ok=True, note=""):
    d = load(); d["total_ops"] += 1
    if not op_ok:
        d["total_fails"] += 1
    if note:
        d.setdefault("patterns", []).append(note)
    json.dump(d, open(MEM, "w", encoding="utf-8"), ensure_ascii=False, indent=2)
    return d

Technical Analysis

record() accepts an arbitrary string in note and appends it directly to persistent JSON state. There is no schema restricting notes to inert observations, no maximum length, no filtering of instruction-like content, no provenance or trust metadata, and no user or project isolation.

The storage operation alone does not execute the supplied text. However, the skill explicitly instructs future agent runs to adopt stored ...[truncated 2277 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the instruction to rewrite SKILL.md automatically. Treat packaged skill instructions as immutable during normal execution.
  2. Store observations in a strictly validated schema with separate fields for event type, capability identifier, timestamp, outcome, and bounded descriptive text.
  3. Never interpret stored free-form notes as executable instructions, policy, or trusted preferences.
  4. Maintain an explicit allowlist of supported preference keys and acceptable values. Reject unknown keys and instruction-like values.
  5. Require explicit, informed user approval before a stored preference can affect later behavior.
  6. Add provenance and trust metadata to every memory entry, including the originating user, task, session, and whether the entry was reviewed.
  7. Isolate memory by user and project instead of using one fixed file shared by every invocation.
  8. Add retention limits, size limits, expiration, inspection, and deletion controls.
  9. If improvements to skill instructions are desired, generate a proposed patch for human review rather than modifying SKILL.md directly.
  10. Use atomic, permission-restricted file writes and fail safely when memory content does not match the expected schema.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The declared description promises a sophisticated meta-agent capability stack: self-verification, self-reflection, orchestration, and an ongoing self-improvement loop that exceeds a teacher skill. The supplied code does not implement those behaviors. It only persists basic operational telemetry to a local JSON file: total operations, total failures, and optional notes. While this logging could be a small supporting component for a larger 'meta-evolver' loop, the provided chunk’s actual primary purpose is bookkeeping, not orchestration or autonomous self-improvement. Therefore the code chunk materially under-delivers relative to the declared purpose, making the description inaccurate for this code.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
92% confidence
Finding

The skill directs use of scripts/learner.py to write persistent state such as learned_patterns.json, but the manifest declares no explicit tool scope or permissions. That creates a capability/transparency gap: operators and downstream systems cannot tell from the metadata that the skill performs file writes and retention. In a public meta-skill that emphasizes autonomous learning and memory, undeclared write capability increases the chance of unintended persistence or misuse.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill is framed as a bootstrap/meta-orchestration utility, yet it stores operation counts, error patterns, user preferences, and improvement suggestions across runs. Persistent tracking is not strictly necessary for the core orchestration purpose and expands data collection beyond user expectations, creating privacy and retention risk. Because the skill is public and marketed as self-evolving, this context makes the extra tracking more concerning, not less.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill states that after every use it automatically reviews activity and accumulates experience, and that learned_patterns.json records usage counts, error patterns, user preferences, and improvement suggestions. Storing this data without an explicit privacy warning or consent mechanism can expose sensitive behavioral data and preferences, especially when retained across runs. In a public skill with self-evolution claims, silent persistence materially increases privacy risk.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs persistent collection and reuse of user preferences and operational history in learned_patterns.json across runs. This creates data-retention and disclosure risk because preferences, repeated errors, and usage patterns can reveal sensitive user traits or work context, and no retention limit, access control, or minimization policy is described. The self-evolving/meta-agent context makes it more dangerous because persistence is central to the workflow and likely to occur repeatedly.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The example command records 输出语言 with the value 中文, which presents a fixed language preference in the skill's natural-language instructions. Because no user choice or opt-in is offered nearby, this can be interpreted as a language policy constraint rather than a selectable example.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

Line 53 states that the learner automatically performs post-use review and accumulation each time the skill is used, implying built-in automatic behavior. But lines 58-70 instruct the operator to manually call python scripts/learner.py ... commands to record success, failure, preferences, and reflections, which contradicts the claimed automatic operation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file presents all user-facing content in a single language, which may violate a language/locale policy if users are not given a choice or informed of the constraint. The file does not document that the skill is region-specific or that Chinese is an intentional, justified locale restriction.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The module docstring is written only in Chinese and presents the skill purpose in that language, which can indicate a fixed language assumption without offering user choice. Under the language/locale policy, natural-language instructions should not force a specific language unless the constraint is clearly documented and justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
85% confidence
Finding

This code file performs a file write to a persistent JSON store, but there is no confirmation prompt, visible log message, or inline warning to inform the user that invocation will modify local state. The module docstring mentions recording calls, but the runtime behavior does not disclose the write when it happens.

Content

No source excerpt is available for this finding.