Back to skill

Security audit

AutoSkill Local Skill Manager

Security checks for vulnerabilities and agentic risk

Overview

This skill is not plainly malicious, but it needs Review because it proactively analyzes sessions and manages durable agent instructions while also allowing unpinned external discovery commands.

Install only if you are comfortable with an agent reviewing sessions for reusable behavior and proposing persistent skill changes. Before use, pin or avoid the `npx skills` commands, keep external discovery disabled for private work unless you approve the exact query and destination, and carefully review every proposed skill diff before approving writes.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
Findings (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:299
Finding

Unpinned Third-Party CLI Execution Through npx

Content
View full analysis

Vulnerability Details

File Location: README.md:19-28; SKILL.md:299-314
Vulnerability Type: Unpinned third-party dependency execution
Risk Level: Medium

Vulnerable Code

README.md:19-28:

bash
npx skills add ECNU-ICALK/autoskill-skill
bash
npx skills add ECNU-ICALK/autoskill-skill --skill autoskill

SKILL.md:299-314:

text
5. When external ecosystem search is requested, duplication risk is high, or the candidate is meant to be public, check well-known sources before creating a new skill. For personal skills, external search is evidence for avoiding duplicates, not a popularity test:
   - browse `https://skills.sh/` or its leaderboard for popular/battle-tested options
   - run `npx skills find <query>` with the best 1-2 queries
   - try alternate terms when the first query misses, e.g. `deploy` vs `deployment` or `pr review` vs `code review`
6. Verify external candidates before recommending them: prefer high install counts, reputable sources, and active source repositories; be cautious with low-install or unknown-author skills.
7. Present external matches with name, purpose, source/quality signal, link, and install command. Do not install, import, overwrite, or merge external skills without explicit user consent.

Technical Analysis

The documented commands invoke the skills npm CLI through npx without pinning a reviewed package version or validating package integrity. If the package is absent from the local cache, npx can dynamically resolve, download, and execute code from the npm ecosystem.

The repository therefore does not completely determine the code executed by these commands. A future package release, compromised maintainer account, registry compromise, dependency confusion event, or malicious transitive dependency could alter the effective executable payload after this skill has been reviewed.

The instruction to verify external skill candida ...[truncated 1625 chars]

Remediation
View remediation

Remediation Suggestions

  1. Pin the CLI to a specifically reviewed version, for example npx skills@X.Y.Z, after confirming the correct package identity and trusted publisher.
  2. Prefer a lockfile-controlled installation with integrity metadata over dynamically resolving a package on every invocation.
  3. Document the expected npm package name, publisher, source repository, version, and checksum or provenance-verification procedure.
  4. Require explicit user confirmation before the first invocation of any dynamically downloaded CLI, including discovery-only commands.
  5. Run the CLI in a restricted environment with minimal filesystem permissions, sanitized environment variables, and no unnecessary credentials.
  6. Disable lifecycle scripts where compatible with the trusted package workflow, and review the CLI plus its dependency tree before upgrades.
  7. Separate approval to execute the discovery CLI from approval to install any skill returned by that CLI.

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:299
Finding

Potential Disclosure of Session-Derived Information Through External Searches

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:299-314, with the storage-redaction rule at SKILL.md:213
Vulnerability Type: Insufficient sensitivity controls for externally transmitted search queries
Risk Level: Medium

Vulnerable Code

SKILL.md:299-314:

text
## Similar Skill Search

Search before creating anything.

1. Identify the candidate's domain, task, output type, tools, and likely synonyms. Build 2-3 short queries such as `<domain> <task>`, `<tool> <output>`, and `<failure-mode> workflow`.
2. Search local skills first:

```bash
rg --files -g 'SKILL.md' <skill-root>
rg -n "<keyword|task|output-type>" <skill-root>
  1. Inspect likely local matches: frontmatter name/description, workflow, triggers, resources, and success criteria. Compare by task family, trigger, tools, failure mode, output contract, and audience.
  2. If a local skill-finder, find-skills, or equivalent discovery tool is installed, use it to broaden the local/external search. Treat results as similarity evidence, not an automatic merge/install decision.
  3. When external ecosystem search is requested, duplication risk is high, or the candidate is meant to be public, check well-known sources before creating a new skill. For personal skills, external search is evidence for avoiding duplicates, not a popularity test:
    • browse https://skills.sh/ or its leaderboard for popular/battle-tested options
    • run npx skills find <query> with the best 1-2 queries
    • try alternate terms when the first query misses, e.g. deploy vs deployment or pr review vs code review
  4. Verify external candidates before recommending them: prefer high install counts, reputable sources, and active source repositories; be cautious with low-install or unknown-author skills.
  5. Present external matches with name, purpose, source/quality signal, link, and install command. Do n ...[truncated 2681 chars]
Remediation
View remediation

Remediation Suggestions

  1. Make local-only discovery the default and require explicit user approval before each external search involving session- or workspace-derived information.
  2. Add a mandatory pre-transmission sensitivity review covering secrets, credentials, personal data, customer identifiers, private URLs, internal project names, and proprietary terminology.
  3. Extend the redaction rule at SKILL.md:213 so it explicitly applies to network requests, CLI arguments, URLs, telemetry, and external search queries.
  4. Convert queries to generic capability terms whenever specificity is unnecessary; for example, replace an internal tool name with its general function.
  5. Show the exact proposed query and destination to the user before transmitting it when the source material is private or ambiguous.
  6. Skip external discovery entirely when de-identification would remove the useful substance of the query.
  7. Document which discovery tools contact external services and avoid passing confidential values through command-line arguments, which may also be retained in shell history or process listings.
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 (21)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 78)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 82)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 108)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 303)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 420)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 493)May include surrounding context.

md
proval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.

Self-Modification

High
Category
Rogue Agent
Confidence
97% confidence
Finding

This skill is designed to edit local skill files, which are durable instructions that can influence future agent behavior and may include scripts or integrations. Even with approval gates, a self-modifying skill manager materially increases the risk of persistence abuse, policy drift, or insertion of unsafe instructions into future execution paths.

Content

Scanner excerpt · SKILL.md (reported line 58)May include surrounding context.

md
- Otherwise check common roots such as `${CODEX_HOME}/skills`, `~/.codex/skills`, `./skills`, `.agents/skills`, and any current workspace skill directory containing `*/SKILL.md`.
- If exactly one plausible root exists, use it.
- If multiple plausible roots exist and the target is ambiguous, ask the user which root to maintain.
- Never reorganize unrelated skill roots or edit skills outside the chosen root.

## Trigger Policy

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Rp1

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README describes proactive detection and extraction checks at 'natural pauses, task boundaries, and session-end reviews,' which are broad triggers that can cause the skill to activate unexpectedly. In a skill manager that inspects session content for reusable material, ambiguous invocation boundaries increase the chance of over-collection, privacy leakage, or unintended skill drafting from sensitive context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill states that output language is determined automatically from the source interaction language and target skill language, without explicit user choice. That can create unsafe behavior if language switching affects user comprehension of prompts, approvals, diffs, or sensitive file operations, increasing the risk of mistaken consent or unnoticed content changes.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
93% confidence
Finding

The skill explicitly aims to preserve session-derived behavior in local SKILL.md folders as a self-improving memory system. That creates durable session persistence, which can store sensitive preferences, context, or mistaken inferences and then affect future interactions beyond the original conversation.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: autoskill
description: Manage personal local Agent Skill files as an installable skill manager. Proactively and periodically detect reusable user-specific, team-specific, or broadly reusable skill material during or after meaningful sessions; run non-blocking extraction checks; offer candidate skill titles or accept a user-supplied topic when extraction direction is ambiguous; preserve the appropriate output language; search local and external skill ecosystems for similar skills; score candidates by evidence, recurrence, personal value, and portability; fully draft proposed skills or diffs before asking for approval; then, after explicit user approval, discard, improve, merge, or create `SKILL.md` folders.
---

# Local Skill File Manager

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The top-level description tells the agent to proactively and periodically detect reusable material across ordinary sessions, which creates an overly broad trigger surface for a persistence-capable skill. In context, this increases the chance the skill activates on unrelated conversations, collects session content, and proposes durable behavioral changes far beyond narrow user intent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger policy directs scanning after each substantive user turn, an ambiguous condition that effectively authorizes continual monitoring of conversation content for extractable patterns. In a self-modifying skill manager, that broadens surveillance and persistence behavior, making accidental capture of sensitive preferences or over-eager skill creation more likely.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
80% confidence
Finding

Telling the model not to ask 'should I check for a skill?' encourages unsolicited analysis of user conversations for reusable patterns. In this skill's context, that reduces user awareness and consent around behavioral profiling and persistence-oriented processing, even if final writes still require approval.

Content

Scanner excerpt · SKILL.md (reported line 68)May include surrounding context.

md
2. Extraction check: when a trigger below fires, classify the experience, assign evidence and value, search for similar skills, and test it against the extraction boundary.
3. Write/update: only edit local skill files after the candidate passes the boundary, the target skill root is clear, and the user approves the exact change.

The model should proactively initiate scans and extraction checks. Do not ask the user "should I check for a skill?" Ask only when the target root is ambiguous, the edit would overwrite uncertain behavior, or the user must approve an external install or sensitive script.

## Candidate Lifecycle

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 121)May include surrounding context.

md
- every file that would be added, modified, deleted, or left untouched
- validation plan or command to run after approval

Ask for a clear yes/no approval only after showing the full proposed artifact or diff. If the user does not approve, do not edit the skill; report the candidate as deferred or discarded.

## Immediate Triggers

Rp1

Medium
Category
MCP Rug Pull
Confidence
89% confidence
Finding

The skill instructs use of npx skills find <query> without pinning an exact package version or verifying package provenance. That creates a supply-chain risk: a future package update, typo-squat, or compromised upstream release could execute untrusted code on the user's machine during routine skill discovery.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 356)May include surrounding context.

md
Decision outcomes before confirmation:

- `discard`: may be reported without confirmation because no file changes occur.
- `keep_note`: requires confirmation if it writes to any memory file; otherwise may be included in the completion note.
- `improve`, `merge`, `create`: always require confirmation before file changes.