Back to skill

Security audit

learn-from-experience

Security checks for vulnerabilities and agentic risk

Overview

This is a disclosed memory skill, but it writes learned preferences into persistent agent configuration across sessions and products in ways that need review before installation.

Install only if you want a persistent memory system that can change future agent behavior. Review the exact files it will edit, prefer a local-only mode, require a diff and explicit approval before any global config sync, and decline the optional companion skill unless its publisher/version are separately reviewed.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (3)

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:252
Finding

Unvalidated learned content is persisted into auto-loaded global agent configuration

Content
View full analysis
extract ## Confirmed Preferences 2. Compile each entry: "### title | confidence" -> "- [tag] description" 3. Detect agent product -> locate global config file 4. Read global config file 5. Find ### Patterns block (create if missing, under ## Learnings) 6. Replace block content (preserve everything else) 7. Update ~/learn-from-experience/index.md sync timestamp + status ``` ### Agent Detection To find the correct global config file: ``` Check in order (first existing path wins): 1. ~/.claude/CLAUDE.md (Claude Code) 2. ~/.openclaw/AGENTS. ...[truncated 2925 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
setup.md:133
Finding

Setup recommends an unpinned third-party Skill installation

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
setup.md:150
Finding

Installation modifies multiple persistent steering surfaces beyond the local memory store

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (31)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The top-level description advertises permanent improvement and cross-session persistence, including compilation into global config, without a prominent warning that user statements may be stored and propagated across future sessions. This is dangerous because users may reveal preferences, habits, or sensitive information in ordinary conversation that then becomes persistently available through auto-loaded configuration.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The sync protocol copies confirmed preferences from the memory store into the agent's global config so they are automatically loaded in new sessions, expanding the blast radius of retained data. This is especially dangerous because global config is a cross-session, cross-context control plane: natural-language content inserted there can leak sensitive user details and influence future model behavior outside the original context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction explicitly tells the agent to trigger a global config sync based on a state marker, but it provides no requirement for user consent, visibility, or confirmation before modifying cross-session configuration. In a skill whose purpose is persistent self-learning across sessions, silent config writes are especially risky because they can propagate mistakes, prompt-injection artifacts, or unauthorized state changes beyond the current task.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The description encourages activation in broad, common situations such as failures, corrections, or discovering a better approach, which can cause the skill to run frequently without a narrowly scoped trigger. In this skill's context, broad activation matters because activation can lead to persistence of user-derived content and eventual modification of globally loaded config files.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill is explicitly designed to retain user preferences and corrections across sessions and to compile them into globally auto-loaded config, creating a durable natural-language data retention channel. Even absent credentials, plain-language memory can capture sensitive personal, project, or operational details and make them broadly available to later sessions or tools that read the same config.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The 'When to Use' section is ambiguous and permissive, allowing the skill to be invoked for ordinary quality-improvement behavior without constraints. Because the skill writes to persistent local storage and can sync into auto-loaded global config, vague invocation guidance increases the chance of unintended retention and configuration changes.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The automatic logging rules instruct the agent to persist user statements that may contain enduring preferences or personal habits in plain-language files. In context, the examples are broad enough that users could unintentionally disclose sensitive information, and the skill lacks a strong mandatory confirmation step before storing such material.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
87% confidence
Finding

The skill authorizes autonomous promotion, demotion, archiving, and sync-adjacent memory management based on internal heuristics rather than user approval. In this context, autonomous state changes are risky because they can silently alter what is remembered and what gets surfaced to future sessions, even if deletion itself requires asking.

Content

Scanner excerpt · SKILL.md (reported line 209)May include surrounding context.

md
- Pattern used 3x in 7 days -> promote to HOT
- Pattern unused 30 days -> demote to WARM
- Pattern unused 90 days -> archive to COLD
- Never delete without asking

### 4. Namespace Isolation
- Project patterns stay in `projects/{name}.md`

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The stale-state auto-sync behavior authorizes autonomous modification of global config on session start without a prominent warning or fresh consent. In context, this is especially risky because it happens during initialization, when neither the current task nor the user's expectations may justify config changes.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
90% confidence
Finding

The protocol creates or edits a persistent, auto-loaded config block and updates tracking state, establishing durable session persistence beyond the current interaction. In this skill's context, that persistence is security-relevant because it can carry user-derived instructions and preferences into future sessions automatically, potentially outside the user's intended scope.

Content

Scanner excerpt · SKILL.md (reported line 275)May include surrounding context.

  1. Compile each entry: "### title | confidence" -> "- [tag] description"
  2. Detect agent product -> locate global config file
  3. Read global config file
  4. Find ### Patterns block (create if missing, under ## Learnings)
  5. Replace block content (preserve everything else)
  6. Update ~/learn-from-experience/index.md sync timestamp + status
text

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file explicitly describes creating and retaining a persistent log of user corrections, including timestamps, context, and promotion into longer-term memory, but provides no privacy notice, minimization rules, consent mechanism, or handling guidance. In the context of a cross-session memory skill, this increases the risk of storing sensitive user preferences, project details, or personal information beyond the immediate task without adequate safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs creation and ongoing auto-updating of files under the user's home directory, including cross-session persistence via a global config sync, without any consent flow, safety warning, scope limitation, or explanation of privacy implications. In an agent skill, silent persistence is risky because it can store user-derived data across sessions, modify local state unexpectedly, and create durable behavior changes that the user did not clearly authorize.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file states that the agent will load and update a persistent memory file across sessions, but the skill metadata does not clearly warn users that session-derived preferences and patterns may be automatically written to disk for future reuse. This can create privacy, consent, and data-retention risks because user-provided corrections or behavioral patterns may be stored and propagated beyond the current session without explicit awareness or approval.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The instruction to trigger a global config sync when the cross-session status is stale can modify shared persistent state across sessions without any explicit user confirmation or warning. Because this skill is specifically designed to propagate learned state across agent products, an automatic sync can spread incorrect, poisoned, or privacy-sensitive data beyond the current task context.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrase 'Recent changes?' is common natural language and can be matched accidentally in unrelated conversation. Generic triggers raise the risk of unintended disclosure or activation of memory-related behavior without clear user intent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The operational description does not clearly warn that the skill may modify multiple global agent config files. Users cannot give meaningful informed consent if high-impact side effects are omitted from the activation and startup description.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The automatic session-start behavior is broad enough to trigger on ordinary use, causing hidden state loading and potentially follow-on sync actions. Broad triggers increase the chance that unrelated conversations activate persistence and file operations the user did not intend.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
92% confidence
Finding

The skill writes user corrections and patterns into persistent files, enabling cross-session retention of potentially sensitive instructions or data. Persistence is especially risky here because stored preferences can later be promoted and synced into multiple global agent configs, compounding the effect of a mistaken or maliciously induced entry.

Content

Scanner excerpt · operations.md (reported line 37)May include surrounding context.

  • Bump counter, update timestamp
    • If counter >= 3: ask to confirm as rule
  1. Determine namespace (global, domain, project)
  2. Write to appropriate file
  3. Update index.md line counts
  4. If written to Confirmed Preferences -> trigger global config sync
text

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

Weekly maintenance performs file moves, archival, compaction, and possible sync activity, but the description lacks a clear warning about these persistent mutations. Hidden background maintenance can surprise users and alter or remove data beyond the immediate task context.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill enumerates and modifies multiple external agent global configuration files outside its own memory store. That expands scope from local self-reflection into cross-application persistent state modification, creating a confused-deputy risk where ordinary use of the skill can silently alter other agents' future behavior.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The stale-detection flow performs auto-sync on activation, which can cause unsolicited writes merely because the skill was invoked or a session started. Automatic persistent writes without a fresh user action undermine user intent and can propagate incorrect or adversarially induced preferences into global configs.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

Binding auto-sync to generic 'skill activation' leaves boundaries unclear and makes side effects depend on ambiguous activation semantics. In practice this can convert passive use into persistent cross-file modification, especially when combined with broad startup logic.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

These instructions direct modification of global config files under the user's home directory and creation of persistent files without an upfront warning that the setup changes cross-session behavior outside the current workspace. Silent or under-disclosed writes to ~/.claude, ~/.codex, and similar locations can alter future agent behavior and make rollback difficult.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The skill explicitly directs the agent to persist corrections, preferences, workflows, and reusable lessons across sessions, creating a long-lived memory store of user-derived data. Even if intended as a productivity feature, persistent retention can capture sensitive behavioral or project information without clear minimization, review, or deletion controls.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The setup flow for this skill instructs the agent to install and activate an unrelated proactivity skill, expanding capabilities and trust surface beyond what the user requested. Chaining installation of another skill can create consent confusion and introduce additional instructions, persistence, or behaviors that were never independently reviewed.

Content

No source excerpt is available for this finding.